See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/228456835

Evaluation of an ASM 1 model calibration procedure on a municipal-industrial

wastewater treatment plant

Article in Journal of Hydroinformatics · January 2002

DOI: 10.2166/hydro.2002.0003

CITATIONS READS
185 2,281

4 authors, including:

Krist V Gernaey Mogens Henze

Technical University of Denmark Technical University of Denmark
637 PUBLICATIONS 15,706 CITATIONS 147 PUBLICATIONS 16,256 CITATIONS

SEE PROFILE SEE PROFILE

Peter A Vanrolleghem
Université Laval
870 PUBLICATIONS 27,710 CITATIONS

SEE PROFILE

All content following this page was uploaded by Krist V Gernaey on 04 June 2014.

The user has requested enhancement of the downloaded file.

 15 © IWA Publishing 2002 Journal of Hydroinformatics | 04.1 | 2002

Evaluation of an ASM1 model calibration procedure on a

municipal–industrial wastewater treatment plant
Britta Petersen, Krist Gernaey, Mogens Henze and Peter A. Vanrolleghem

ABSTRACT

The purpose of the calibrated model determines how to approach a model calibration, e.g. which Britta Petersen
Peter A. Vanrolleghem
information is needed and to which level of detail the model should be calibrated. A systematic BIOMATH Department,
Ghent University,
model calibration procedure was therefore defined and evaluated for a municipal–industrial Coupure Links 653,
wastewater treatment plant. In the case that was studied it was important to have a detailed B-9000 Gent,
Belgium
description of the process dynamics, since the model was to be used as the basis for optimisation
Krist Gernaey (corresponding author)
scenarios in a later phase. Therefore, a complete model calibration procedure was applied including: BIOMATH Department,
Ghent University,
(1) a description of the hydraulics in the system via a tracer test, (2) an intensive measuring and
Department of Civil, Geological and Mining
campaign and (3) supporting lab-scale experiments to obtain and confirm kinetic parameters for the
Engineering,
model. In this paper the model calibration procedure for this case study is described step by step, École Polytechnique de Montréal.
Current address: CAPEC,
and the importance of the different steps is discussed. The calibrated model was evaluated via a Department of Chemical Engineering,
Technical University of Denmark,
sensitivity analysis on the influence of model parameters and influent component concentrations on Building 229,
DK-2800 Lyngby,
the model output. The sensitivity analysis confirmed that the model output was sensitive to the
Denmark
parameters that were modified from the default parameter values. The calibrated model was finally Tel: +45 45 25 28 00
Fax : +45 45 88 22 58
reduced from a 24 tanks-in-series configuration to a 12 tanks-in-series configuration, resulting in a E-mail: Krist.Gernaey@olivia.kt.dtu.dk

50% reduction of the simulation time. Mogens Henze

Environment and Resources
Key words | nitrification, denitrification, ASM1 model, calibration procedure, sensitivity analysis, Technical University of Denmark,
model reduction Building 115,
DK-2800 Lyngby,
Denmark

INTRODUCTION
Implementation of biological nutrient removal on waste- Group on Mathematical Modelling for Design and
water treatment plants (WWTPs) resulted in an increased Operation of Biological Wastewater Treatment Processes
knowledge of the biological degradation processes. This (Henze et al. 1987) is generally accepted as the state-of-
resulted in the development and use of more advanced the-art. ASM1 was primarily developed for municipal
dynamic mathematical models that may be able to activated sludge wastewater treatment plants to describe
describe the biological nutrient removal processes. These the removal of organic carbon substances and nitrogen
activated sludge models allow us to study and to further with simultaneous consumption of oxygen and nitrate as
increase our understanding of the influence of process electron acceptors, and to yield a good description of the
modifications on treatment process efficiency. The sludge production. ASM1 has been extended to include a
dynamic models are, for example, increasingly used for description of biological phosphorus removal, resulting in
scenario evaluations aiming at the optimisation of acti- ASM2 and ASM2d (Henze et al. 1995, 1999). Recently,
vated sludge processes (Stokes et al. 1993; de la Sota et al. some of the model concepts behind ASM1 have been
1994; Coen et al. 1997, among many others). The Activated altered in ASM3 (Gujer et al. 1999), a model that also
Sludge Model No. 1 (ASM1) presented by the IAWQ Task focuses on the degradation of carbon and nitrogen but
 16 Britta Petersen et al. | Evaluation of an ASM1 model calibration procedure Journal of Hydroinformatics | 04.1 | 2002

allows the introduction of processes describing the storage some intermediate streams between the
of biopolymers under transient conditions. WWTP unit processes), as averages or
In this study model calibration is understood as the dynamic trajectories, e.g. suspended solids
adaptation of the model to fit a certain set of information (SS), chemical oxygen demand (COD),
obtained from the full-scale WWTP under study. This task Kjeldahl nitrogen (TKN), ammonium
is often rather time-consuming, and typically the time nitrogen (NH4-N), nitrate nitrogen
needed for a model calibration is underestimated. Even (NO3-N), orthophosphate (PO4-P), etc.
though more than a decade has passed since the publi- 5.2. Sludge composition, e.g. SS, volatile
cation of ASM1, a fully developed model calibration pro- suspended solids (VSS), COD, nitrogen
cedure has not yet been defined. We have not been able to and/or phosphorus content.
find a complete model calibration report in the literature. 5.3. Reaction kinetics, e.g. growth and decay
There may be many reasons for this. It is important to rates.
realise that the purpose of the model is very much in 5.4. Reaction stoichiometry, e.g. yields.
determining how to approach the calibration, making it
As mentioned above, the required quality and quantity of
difficult to generalise (Henze et al. 1995). Still, considering
the information will depend very much on the purpose of
the wide application of these activated sludge models,
the model. In cases when the model is to be used for
there are surprisingly few references that contain details
educational purposes (e.g. to increase basic understanding
on the applied model calibration procedure. Often one has
of the processes), for comparison of design alternatives for
to collect bits and pieces from various sources to obtain an
non-existing plants or in other situations where qualitative
overview.
comparisons are sufficient, the default parameter values
In this study we have attempted to gather and
defined by Henze et al. (1987) can be applied. A reasonably
summarise the information needed to achieve a successful
good description can most often be obtained with this
model calibration. The set of information listed below was
default parameter set for typical municipal cases without
extracted and combined from different sources (Henze
significant industrial influences (Henze et al. 1997). How-
et al. 1987; Lesouef et al. 1992; Pedersen & Sinkjær 1992;
ever, if the calibrated model is going to be used for process
Siegrist & Tschui 1992; Stokes et al. 1993; de la Sota et al.
performance evaluation and optimisation, it may be
1994; Dupont & Sinkjær 1994; Weijers et al. 1996; Xu &
necessary to have a more accurate description of the
Hultman 1996; Kristensen et al. 1998):
actual processes under study. Some processes may need a
more adequate description than others, again depending
1. Design data, e.g. reactor volume, pump flows and
on the purpose of the study. This may especially apply for
aeration capacities.
models that are supposed to describe the processes in an
2. Operational data:
industrial or combined municipal and industrial treatment
2.1. Flow rates, as averages or dynamic
plant.
trajectories, of influent, effluent, recycle and
The information needed for the characterisation of the
waste flows.
biological model, listed in point 5 above, can basically be
2.2. pH, aeration and temperatures.
gathered from three sources:
3. Characterisation for the hydraulic model, e.g. the
results of tracer tests. 1. Default values from literature (e.g. Henze et al.
4. Characterisation for the settler model, e.g. zone 1987).
settling velocities at different mixed liquor 2. Full-scale plant data
suspended solids concentrations. 2.1. Average or dynamic data from grab or
5. Characterisation for the biological model, ASM1, of: time/flow proportional samples.
5.1. Wastewater concentrations of full-scale 2.2. Conventional mass balances of the full-scale
WWTP influent and effluent (as well as data.
 17 Britta Petersen et al. | Evaluation of an ASM1 model calibration procedure Journal of Hydroinformatics | 04.1 | 2002

settling and biological characterisation of the full-scale

WWTP. The hydraulics (3) are typically characterised via
tracer tests at the full-scale installation. The settling
characteristics (4) can be characterised via on-line or
lab-scale settling tests (Vanderhasselt et al. 1999). Finally,
the biology can be characterised via different information
sources, as indicated above. A review of the information
that can be obtained from different kinds of lab-scale
experiments is presented in detail elsewhere (Petersen
2000), and for information especially obtained from
respirometric tests the reader is referred to Vanrolleghem
et al. (1999).
In Figure 1 steps 6–10 illustrate different calibration
levels. The calibration of the hydraulic model via tracer
test results, and the settler model calibration via results
from sludge settling tests, are indicated in steps 6 and 7,
Figure 1 | Schematic overview of the different general steps in an activated sludge
model calibration procedure.
respectively. A first ASM calibration level is typically a
simple steady state model calibration step (8). In this
phase of the model calibration the different reactors in the
treatment plant are each represented by an ideal perfectly
mixed tank, resulting in a simple treatment plant con-
2.3. On-line data.
figuration. Here data obtained from the full-scale WWTP
2.4. Measurements in reactors to characterise
are averaged, thereby assuming that this average repre-
process dynamics (mainly relevant for
sents a steady state, and the model is calibrated to fit to
sequencing batch reactors (SBRs) and other
average effluent and sludge waste data. Typically, the
alternating systems).
calibration of the ASM and the settler are linked together,
3. Information obtained from different kinds of
since the aim is most often to describe the final effluent
lab-scale experiments with wastewater and activated
quality. Moreover, the recycle from the settler has an
sludge from the full-scale plant under study.
influence on the activated sludge system. Thus, at this
Again, the intended use of the model will determine which stage, there may be an interaction between the steady state
information source to choose for the characterisation of calibration and the settler model calibration, indicated by
the different biological processes in the model. In the double arrow. Finally, the characterisation of waste-
addition, the purpose will decide to which level the model water components may be adjusted according to the cali-
has to be calibrated, since the quality of the desired model bration of the full-scale model, indicated with the double
predictions will depend strongly on the quality of the arrow between (8) and (5) in Figure 1.
model calibration. Figure 1 illustrates the different general The next step in the calibration procedure is a steady
steps in a model calibration procedure. It should be state model calibration that includes the hydraulic model
stressed that, depending on the purpose, not all steps may (9). In general, with a steady state model calibration, only
have to be taken. parameters responsible for the long-term behaviour of the
Steps 1–5 in Figure 1 indicate the collection of WWTP can be determined, i.e. YH, fp, bH and XI in the
information. Design (1) and operational (2) data are in influent (Henze et al. 1999; Nowak et al. 1999). These
general always needed for a model calibration. For parameters are correlated to a certain degree, meaning
example, the flow and load variations are important in the that a modification of one parameter value can be com-
design of measuring campaigns for hydraulic, sludge pensated by a modification of another parameter value. In
 18 Britta Petersen et al. | Evaluation of an ASM1 model calibration procedure Journal of Hydroinformatics | 04.1 | 2002

the study of Nowak et al. (1999) on mass balances of (Ljung 1987). However, since measurements on full-scale
full-scale data, it was therefore chosen to fix YH and fp, WWTPs are relatively expensive these recommendations
leaving XI in the influent and bH to be determined from may not always be completely fulfilled.
the steady state data. In the study of Lesouef et al. (1992) Furthermore, data from the full-scale installation
two WWTP models were calibrated via steady state cali- alone may be insufficient for a dynamic model calibration
bration only, and this calibrated model was applied to since the reaction kinetics cannot readily be obtained
simulate dynamic process scenarios. However, if one from such data, except for specific designs like SBRs and
relies entirely on a steady state calibration to dynamic alternating systems (Vanrolleghem & Coen 1995). For a
data, some problems may be encountered since the real dynamic model calibration on a full-scale WWTP the
input variations are usually faster than the slow process modeller is therefore typically aiming at combining more
dynamics that were focused upon during the steady state information-rich results derived from lab-scale exper-
calibration. In other words, the process does not operate iments (carried out with sludge and wastewater from the
in steady state but one still attempts to fit a steady state full-scale installation) with data obtained from measuring
simplification of the model to an unsteady situation. A campaigns on the WWTP under study (Dupont & Sinkjær
steady state calibration may, however, be very useful for 1994; Xu & Hultman 1996; Kristensen et al. 1998).
the determination of initial conditions prior to a dynamic In this paper the model calibration procedure pre-
model calibration and for the initiation of a first par- sented in Figure 1 is concretised for the combined
ameter estimation (e.g. Pedersen & Sinkjær 1992; Stokes municipal–industrial activated sludge WWTP in Zele,
et al. 1993; Dupont & Sinkjær 1994; Xu & Hultman 1996; Belgium. The purpose of the model calibration was to
Kristensen et al. 1998). obtain a good description of the nitrogen removal capacity
If it is the aim to describe and predict more short-term and, to a lesser extent, of the COD removal. In a second
and dynamic situations, a model calibration to dynamic phase the model was to be applied for process optimis-
data will be needed since such data contain more ation of nitrogen removal (Gernaey et al. 2000). Based on
information than steady state data, especially on fast this purpose the necessary information set and calibration
dynamic behaviour. The important point in model strategy were defined. A tracer test was carried out first,
calibration based on dynamic data is to obtain a more to have an adequate description of the hydraulic flow
reliable estimation of the maximum specific growth pattern, which is especially important if dynamic situ-
rates mmaxH and mmaxA (Henze et al. 1999), which are ations are to be predicted. The sludge at the Zele WWTP is
the most important parameters in predicting dynamic settling reasonably well, and it was therefore found
situations. adequate to describe the settler with a simple point-settler
At WWTPs, data are most often collected routinely at model. Thus, no specific tests were carried out to charac-
a daily or weekly sampling frequency. This sampling fre- terise the settling properties, i.e. step 4 of Figure 1 is not
quency may, however, not be high enough, and for more included in this study. With respect to the wastewater
accurate modelling it may therefore be required to run characterisation an intensive measuring campaign was
special measuring campaigns (e.g. Pedersen & Sinkjær designed to obtain sufficient dynamic data. The variation
1992; de la Sota et al. 1994; Dupont & Sinkjær 1994; Xu & in readily biodegradable organic substrate was character-
Hultman 1996; Coen et al. 1997). The sampling frequencies ised, since the model was to be applied later on for
should be chosen in relation to the time constants of the optimisation of nitrogen removal, including the start-up of
process and influent variations. One of the important time denitrification. The sludge composition was analysed to
constants of the process is the hydraulic retention time. support the calibration of the sludge balance. Moreover,
Various lengths of measuring campaigns are recorded in lab-scale experiments were planned for the determination
the literature. Ideally, one should choose to sample about of the sludge kinetics related to nitrification and COD
five times faster than the hydraulic retention time and degradation, and a decay experiment was carried out to
have a test duration of 3–4 times this key time constant support the description of biomass decay in the treatment
 19 Britta Petersen et al. | Evaluation of an ASM1 model calibration procedure Journal of Hydroinformatics | 04.1 | 2002

aerated channel of about 200 m3. The clarifiers each have a

diameter of 33 m and a volume of 2050 m3. The final efflu-
ent is discharged into a nearby stream. The underflow from
the secondary clarifier flows back to the aeration tank
through an aerated sludge recycle channel with a volume of
400 m3. The primary and secondary sludge are thickened
prior to anaerobic digestion.
The Zele WWTP is going through a stepwise reno-
vation process with the aim of obtaining an effluent qual-
ity that complies with the Flemish effluent standard for
total nitrogen (15 mg total nitrogen per litre as a yearly
average). A first step in the renovation process was the
Figure 2 | Schematic overview of the process layout of the Zele wastewater treatment installation of a fine bubble aeration system in 1997. This
plant (Aquafin NV, Aartselaar, Belgium).
had an immediate positive effect on the nitrification
capacity, resulting in a decrease of the effluent NH4-N
concentration. However, the effluent total nitrogen con-
plant. No specific experiments were carried out for the centration remained too high to comply with the effluent
determination of stoichiometric coefficients. A sensitivity standard. An average total nitrogen concentration of
analysis was carried out on the calibrated model to check 19.6 mg/l was calculated for the period January 1997 to
whether the parameters that were modified during the November 1998 (the average of 53 effluent samples). The
model calibration procedure were indeed influencing the absence of a denitrification compartment in the WWTP is
model output. It was finally investigated if the calibrated believed to be the main reason for the high effluent total
model could be reduced, to increase simulation speed nitrogen concentrations.
while maintaining the same accuracy of the full model.

MATERIALS AND METHODS

CASE STUDY
Tracer test
The municipal activated sludge WWTP at Zele (Aquafin
NV, Aartselaar, Belgium) was constructed in 1983 for a A tracer test with lithium chloride (LiCl) was carried out
design capacity of 50,000 inhabitant equivalents (IE). Fig- to characterise the hydraulics of the activated sludge tank.
ure 2 gives a schematic overview of the process layout. The The tracer was added as a pulse at the beginning of the
influent of the WWTP consists of 40% household waste- aeration tank, where presettled influent is mixed with
water and 60% industrial wastewater (slaughterhouses, in- return sludge (indicated in Figure 2). During the test
dustrial laundries, textile cleaning, textile painting, etc.). mixed liquor samples were taken at the point where the
The influent is divided over two parallel rectangular primary activated sludge flows over into the secondary clarifiers
clarifiers after the pretreatment step (coarse grit removal, (see Figure 2). The sampling frequency took into account
fine grit removal, sand and grease removal). The effluent of the worst case scenario (with respect to obtaining detect-
the primary clarifier flows to the biological activated sludge able Li concentrations) of an ideally mixed situation,
treatment, where it is mixed with recycled sludge. The acti- although the expectation in view of the design was plug-
vated sludge tank consists of one plug-flow aeration tank flow mixing behaviour. Thus, frequent sampling (one
that is divided into 6 lanes of about 400 m3 each. The mixed sample every 5–10 minutes) was undertaken for 0–1.5
liquor flows to two secondary clarifiers through an open times the hydraulic retention time.
 20 Britta Petersen et al. | Evaluation of an ASM1 model calibration procedure Journal of Hydroinformatics | 04.1 | 2002

Measuring campaign biodegradable COD (CODst) of the influent. The BODst

was determined as the area under the oxygen uptake rate
The measuring campaign was carried out in November
profile related to substrate degradation (a respirogram).
1998. First, a 1-day campaign was done to test the planned
Only unfiltered influent samples were subjected to respiro-
strategy, e.g. to evaluate whether the planned measure-
metric analysis. Some of the samples (typically the ones
ment frequency of one sample every two hours was high
with a high COD concentration) were also analysed after
enough to observe the dynamics, and to check if the
inhibiting nitrification with allylthiourea (ATU). The
installed measuring/sampling equipment worked prop-
measured BODst values were converted to COD units via
erly. Only afterwards was a detailed one-week measuring
an assumed yield factor YH of 0.67 (Henze et al. 1987):
campaign carried out. Two automatic samplers with
built-in refrigerator (4°C) were installed on the treatment
plant. Time proportional samples (100 ml every 6 min)
were taken every second hour on the effluent of the pri-
mary clarifier ( = influent to activated sludge tank) and on For the case when BODst was available from experiments
the effluent of the secondary clarifier (see Figure 2). The in the presence of ATU, the value was immediately used as
influent samples were analysed for the following par- an estimate for the CODst concentration. For the case
ameters via standard methods: SS, NH4-N, TKN, total and when the BODst value resulted from an experiment in
soluble chemical oxygen demand (CODtot and CODsol). which no ATU was added, the CODst concentration was
The samples for CODsol were filtered through 0.45 µm determined according to (2). The BODst requirement for
filters prior to analysis. The effluent samples were analysed the oxidation of NH4-N (BODst,NH4) was determined
for SS, NH4-N, NO3-N, nitrite nitrogen (NO2-N), CODtot using the NH4-N concentrations obtained from the
and CODsol. In addition, mixed liquor and return sludge chemical analyses of the wastewater (3). The value of YA
were sampled at regular times (once or twice per day) to was set to 0.24 (Henze et al. 1987), thereby neglecting the
measure the COD, TKN, SS and VSS content of the sludge. amount of NH4-N consumed for biomass growth during
Effluent flow data were collected with a data logger the tests. The latter is a reasonable and common assump-
that was temporarily connected to the effluent flow tion in short-term respirometric experiments such as the
sensor. Sludge waste flows were obtained from treatment ones carried out in this study:
plant operation logbooks. Temperature and pH of the
activated sludge in the aeration tank were measured daily. BODst = BODst,total − BODst,NH4 (2)

BODst,NH4 = (4.57 − YA)NH4 − N (3)

Lab-scale experiments

Respirometric lab-scale experiments with wastewater and

Activated sludge kinetics—maximum specific growth
activated sludge were carried out during the measuring
rates
campaign. Experiments were performed in the RODTOX
(Vanrolleghem et al. 1994) and the hybrid respirometer Experiments were carried out to obtain data to estimate
(Gernaey et al. 2001). The purposes of these experiments the kinetic parameters related to nitrification. The design
were two-fold. of these experiments is described in more detail elsewhere
(Petersen 2000), but consisted of simultaneous addition of
wastewater and ammonium, thus allowing us to estimate
Wastewater characterisation—determination of readily
the nitrification kinetics and the degradation of COD in a
biodegradable COD
single experiment. The exogenous oxygen uptake rate,
The first purpose was to measure the short-term biochemi- rO,ex, caused by the wastewater and ammonium addition
cal oxygen demand (BODst) to determine the readily can be described by (4). Note that in (4) one biomass X is
 21 Britta Petersen et al. | Evaluation of an ASM1 model calibration procedure Journal of Hydroinformatics | 04.1 | 2002

used for the interpretation of respirometric data derived Table 1 | Design data from Zele WWTP

from short-term respirometric tests. This is done on Design parameter Unit Value
purpose, since it is not known which fraction of the
particulate COD corresponds to XBH and XBA: Volume activated sludge tank m3 2,600

Volume of recycle channel m3 400

Volume of secondary clarifier m3 2 × 2,050

Decay rate the first day of the measuring campaign, as can be seen
from the flow data (2 hour averages) in Figure 3. Therefore
The endogenous respiration rate, rO,end, was measured as a some key parameters were calculated both including
function of time in a long term (5 days) aerated batch and excluding the data obtained during this rain event
experiment without substrate supply. The observed en- (Table 2). Table 2 clearly shows that the daily COD load,
dogenous decay rate, b′H, was determined as the slope of and thereby the sludge load, increased significantly during
the curve consisting of ln(rO,end(0))/ln(rO,end(t)) data the rain period. The sludge age seems low for a nitrifying
points plotted as a function of time (Ekama et al. 1986). WWTP. It is the experience at the WWTP, however, that it
This decay rate was transformed into the model decay rate is difficult to maintain a higher sludge age during winter
based on the death regeneration concept via (5) (Henze due to a decrease in sludge settleability and thereby an
et al. 1987), where YH was set equal to 0.67 and fp to 0.08 increased risk for sludge washout. The observed yield is
according to the ASM1 default parameters also slightly higher than expected according to the sludge
load (Henze et al. 1997) but is probably related to the low
sludge age (young sludge).

Steps 3 and 6: characterisation of hydraulics and

Temperature correction of the parameters determined
calibration of hydraulic model
from the lab-scale experiments (18°C) was carried out
according to standard procedures (Henze et al. 1997). The data resulting from the tracer test are shown in
Figure 4. A sharp peak was recorded with a maximum Li
concentration of 1.2 mg/l at t = 0.1 d. The increase of the
Li concentration around t = 0.27 d is due to the Li that is
recycled internally in the treatment plant with the sludge
RESULTS recycle.
The tracer test data were normalised by CO (6), where
The results of the different model calibration steps,
M is the total mass of Li added at t = 0, and the time was
outlined in Figure 1, are described for the example that
normalised by the average hydraulic residence time (OH)
was studied.
during the test, which was 2.6 h. First, a simple data
interpretation was applied. The N tanks-in-series model
(7) was fitted to the normalised Li data via the solver
Steps 1–2: design and operational data
function in MS Excel and N = 19 gave the best fit:
The volumes are repeated in Table 1, and Table 2 lists the
operational data during the 6 day measuring campaign
(18–23 November 1998). The data include a rain event on
 22 Britta Petersen et al. | Evaluation of an ASM1 model calibration procedure Journal of Hydroinformatics | 04.1 | 2002

Table 2 | Operational data from Zele WWTP obtained during the measuring campaign

Variable Unit Value (incl. rain period) Value (excl. rain period)

Influent flow average m3/d 12,559 10,255

Waste flow average m3/d 248 241

Temperature °C 10.5 10.5

pH 7.2 7.2

Sludge concentration average* g SS/l 4.0 3.9

COD load kg COD/d 5,607 3,730

TKN load kg TKN/d 342 258

Sludge load kg COD/kg SS. d 0.48 0.32

Sludge production kg SS/d 2,394 2,300

Sludge age d 6.2 8.6

Observed yield kg SS/kg COD 0.42 0.62

*See also Table 3.

This simple approach, however, assumes a constant flow

rate and thus does not include the flow variations that
occurred during the tracer test. Moreover, the sludge re-
cycle (including the LiCl that is recycled via the underflow
of the settlers) and the residence time of the recycle liquid
flow in the secondary clarifiers were not considered in this
simplified approach. Thus, to obtain a better hydraulic
description different configurations were simulated in the
WEST + + modelling and simulation environment
(Hemmis NV, Kortrijk, Belgium). The value of N = 19
obtained from the simple data interpretation was used as a
starting point for this more detailed model-based interpret-
ation. The resulting plant configuration that gave the best
description consisted of the following components:

• A plug-flow aeration tank consisting of 24 tanks in

series (6 lanes, each consisting of 4 tanks in series,
Figure 3 | Simulated flow rates during the measuring campaign (t=0 corresponds to the
where each tank has a volume of 100 m3).
beginning of the measuring campaign). Influent flow rates (2 h averages) were
collected at the treatment plant. Settler underflow rates are obtained as
• The sludge channel that transports the sludge from
0.55×influent flow (proportional recycle flow controller). The discontinuous the aeration tank to the secondary clarifiers consists
waste flow rates were obtained from treatment plant operation logbook data. of two tanks in series of 100 m3 each.
 23 Britta Petersen et al. | Evaluation of an ASM1 model calibration procedure Journal of Hydroinformatics | 04.1 | 2002

• An ideal point-settler and a ‘buffer tank’ of 1000 m3

to take the residence time into account for the liquid
that is recycled together with the recycle sludge
(settler underflow).
• The recycle channel that transports the recycle
sludge from the secondary clarifiers to the aeration
tank consists of five tanks in series of 80 m3 each.

The model fit on the Li tracer data for this configuration is

shown in Figure 4, and the hydraulic scheme is illustrated
in Figure 14 in the process configuration for the dynamic
model. Figure 4 | Data resulting from the tracer test on the aeration tank of the Zele WWTP,
together with the best model fit that was obtained. See text for explanation
of the model.

Step 5: biological characterisation

Wastewater characterisation
Figures 5, 6 and 7 show the chemical analysis results for
the influent samples (presettled wastewater). The influent
CODtot, TKN and SS concentrations were highest on
Wednesday afternoon and Thursday morning as a result of
the rain event. The CODsol and NH4-N concentrations
were, however, not higher during the rain event compared
to the other working days, indicating that the increase of
CODtot and TKN concentrations during the rain event
were related to the extra SS load. The high SS content of
the presettled wastewater during the rain event indicates Figure 5 | CODtot and CODsol concentrations measured on the influent of the Zele WWTP
that the primary clarifiers are overloaded when the during the measuring campaign. CODst values were calculated using the
BODst values resulting from respirometric experiments with unfiltered
influent flow is high, e.g. due to rainfall. wastewater (see (1)).

For both COD (Figure 5) and nitrogen (Figure 6) a

diurnal pattern can be distinguished, with lower concen-
trations during the night and higher during daytime. The
concentration variations are much lower at the weekend
due to the absence of industrial discharges. The start-up of
industrial activity after the weekend again caused an
increase of the influent pollutant concentrations. In
Figure 5 it can furthermore be seen that the CODst,
obtained via BODst from respirometric tests (1), is related
to CODsol. Moreover, the CODst seems to be related to the
industrial discharges since there is hardly any CODst
present in the influent during the weekend.
The results of the chemical analyses on the effluent are
shown in Figures 8, 9 and 10. Effluent COD and N
Figure 6 | TKN and NH4-N concencentrations measured on the influent of the Zele
concentrations show a diurnal pattern but, as expected, WWTP during the measuring campaign.
 24 Britta Petersen et al. | Evaluation of an ASM1 model calibration procedure Journal of Hydroinformatics | 04.1 | 2002

Figure 7 | SS concentration measured on the influent of the Zele WWTP during the Figure 9 | Total nitrogen and NH4-N concentrations measured on the effluent of the Zele
measuring campaign. WWTP during the measuring campaign.

Figure 8 | CODtot and CODsol concentrations measured on the effluent of the Zele WWTP Figure 10 | NO3-N concentrations measured on the effluent of the Zele WWTP during
during the measuring campaign. the measuring campaign.

the variations were rather small compared to the concen- concentration (SO) in the incoming wastewater is zero.
tration variations measured in the presettled influent. Furthermore, the conversion of alkalinity (SALK) is
During the measuring campaign the effluent total nitrogen not considered in this study, since influent SALK is not
concentration was, in general, lower than the 15 mg N/l inhibiting nitrification at the treatment plant. The plant
effluent standard (Figure 9). However, on Monday, a day indeed has no problems in reaching full nitrification dur-
with normal dry weather flow, the effluent total nitrogen ing summer time, when the temperatures are considerably
concentration exceeded the 15 mg N/l standard. Also, in higher compared to winter. The ASM1 wastewater com-
Figure 9 it can be seen that full nitrification was not ponents to consider in the wastewater characterisation are
reached. Besides NH4-N, NO3-N contributed significantly thus related to the organic carbon (COD) and nitrogen
to the effluent total nitrogen concentrations during the components.
measuring campaign (Figure 10).
The next step in the wastewater characterisation is
COD components
the conversion of the available data from the measuring
campaign into a data set that can be used as input for the The total COD in the model includes the components
ASM (in this case ASM1). It is assumed that the oxygen described in (8) (Henze et al. 1987):
 25 Britta Petersen et al. | Evaluation of an ASM1 model calibration procedure Journal of Hydroinformatics | 04.1 | 2002

(available from samples that are routinely taken on

the effluent of the WWTP) were used to determine
the effluent SS concentration. An average effluent
BOD5 concentration of 6.2 ± 3.0 mg/l had been
measured, and an average BOD5/CODtot ratio of
7 ± 3% was calculated for the effluent data.
Assuming a BOD yield (Y) of 0.20 (STOWA 1996),
the corresponding effluent SS concentrations were
calculated according to (9). The choice of this low
BOD yield is due to the fact that significant biomass
decay takes place during a BOD measurement. The
influent SI was obtained by combining (9) and (10):

Figure 11 | Summary of influent characterisation methods for organic wastewater

components.

SI,effluent = SI,influent
= (CODsol.effluent − SS,effluent) (10)

CODtot = SS + SI + XS + XI (8) • Readily biodegradable substrate (SS), influent.

Based on the results of the respirometric tests, the
The presence of heterotrophic and autotrophic biomass influent readily biodegradable COD (1) is set equal
(XBH and XBA) in the influent wastewater was not to the model component SS.
considered in the ASM1 report (Henze et al. 1987). • In the next step, the mass balance for influent
Activated sludge may, however, be inoculated significantly CODsol (CODsol,influent) was checked (11):
by XBH in the influent, especially in cases where no
CODsol,influent = SS + SI + Srest (11)
primary settling is present. However, it can be difficult to
determine the amount of biomass in the wastewater, and
For the case when CODsol,influent in (11) is higher
the biomass fraction is therefore often lumped into XS
than SS + SI, Srest can be added to the slowly
(Henze et al. 1995). This does not influence the modelling
biodegradable substrate, XS. In contrast, if
significantly but it may affect the value of the biomass
CODsol,influent is lower than SS + SI, part of the
yield. Contrary to the heterotrophic biomass, the presence
measured BODst may be considered to originate
of autotrophic biomass (XBA) can be important to keep
from XS. For the wastewater under study it appeared
sufficient nitrification in the system in cases where the
that CODsol,influent > (SS + SI); thus Srest was added to
sludge retention time is too low to sustain the nitrifying
the XS component.
biomass. Model results may reveal whether this is the case.
As an initial approximation XBH and XBA were assumed to
• Slowly biodegradable substrate (XS), influent.
Contribution to the XS concentration partly came
be zero in this study. The methods for characterisation of
from the mass balance in (11) as Srest, but was also
the organic wastewater components are summarised in
partly determined from the steady state model
Figure 11.
evaluations (see also XI).
• Inert soluble organic matter (SI), influent and • Inert suspended organic matter (XI), influent.
effluent. The best estimate for XI is obtained by comparing
Influent SI was determined via effluent data (Henze the measured and predicted sludge concentration
1992). The weekly BOD5 results of the effluent and sludge production (Henze et al. 1987, 1995). The
 26 Britta Petersen et al. | Evaluation of an ASM1 model calibration procedure Journal of Hydroinformatics | 04.1 | 2002

Table 3 | Analysis of results on activated sludge and recycle sludge (average and 95% confidence interval, resulting from 10 measurements)

SS (g/l) VSS/SS COD/SS COD/VSS TKN/COD (%)

Activated sludge 4.01 ± 1.20 0.70 ± 0.02 1.38 ± 0.26 1.99 ± 0.36 3.90 ± 1.31

Recycle sludge 10.05 ± 5.27 0.69 ± 0.02 1.37 ± 0.12 1.98 ± 0.17 3.45 ± 1.38

XI influent concentration is typically used as a (iNSI) equals 1.5% (Henze et al. 1995) the
‘tuning component’ in the model calibration of the concentration of SND can be determined via (15):
sludge balance (Henze et al. 1995; Nowak et al.
1999), and the XS concentration is adjusted
accordingly, via the mass balance of (12) (assuming
that influent XBH and XBA are negligible, see above): SND = SKN − iNSISI − SNH (15)

CODtot − CODsol = XS + X1 (12) • Slowly biodegradable organic nitrogen (XND).

The nitrogen content of inert suspended organic
Initially XI was assumed to be 50% of the particulate matter (iNXI) is initially assumed to be 1% (Henze
COD. However, this was adjusted during the model et al. 1995) resulting in (16) for the determination of
calibration (see below). XND:

XND = TKN − iNXIXI − SKN. (16)

Nitrogen components

For the nitrogen fractions a similar approach was used for Sludge composition
both influent and effluent characterisation. It was
The average results of the sludge composition analysis,
assumed that the influent contains negligible concen-
based on ten measurements of the COD, SS and VSS
trations of nitrate (SNO). The total Kjeldahl nitrogen could
content of the activated sludge and recycle sludge, are
then be fractionated according to (13) (Henze et al.
given in Table 3. The measured COD/VSS ratio is slightly
1987):
higher than typical values, although Stokes et al. (1993)
observed similar values. The high content of reduced
TKN = XNI + XND + SNI + SND + SNH (13)
matter could be due to industrial discharges (e.g. discharge
• Ammonia nitrogen (SNH). of fat from the slaughterhouses).

The analytically measured NH4-N concentration was con-

sidered to be equal to SNH. Kinetic characterisation

• Soluble biodegradable organic nitrogen (SND). In Figure 12 a typical respirogram of a wastewater and a
Only TKN measurements were available. It was respirogram obtained after addition of wastewater plus
assumed that the ratio of soluble to total TKN was extra ammonium are illustrated. It is obvious that the
proportional with the ratio of CODsol to CODtot. wastewater respirogram cannot be separated clearly into
Thus, the soluble Kjeldahl nitrogen (SKN) can be two parts, i.e. one part that describes the oxygen consump-
approximated via (14), and by assuming that the tion due to COD degradation and one part that describes
nitrogen content of inert soluble organic matter the nitrification. Thus, the wastewater respirogram alone
 27 Britta Petersen et al. | Evaluation of an ASM1 model calibration procedure Journal of Hydroinformatics | 04.1 | 2002

Figure 13 | Treatment plant configuration used for the steady state model calibration.

(V = 400 m3), and a constant average sludge waste flow

from the recycle line.
As mentioned above, the main aim of the steady state
Figure 12 | Example of respirograms obtained from respirometric tests with
wastewater and wastewater mixed with ammonium. model is to fit the modelled sludge production to the
sludge production calculated from plant data collected
during the measuring campaign (based on waste flow data
and sludge concentration measurements). This is done by
adjusting parameters responsible for long-term behaviour,
is not informative enough for the identification of both the
i.e. the decay rates bH and bA, together with the influent
nitrification kinetics and the degradation kinetics related
concentration of XI. Furthermore, for the point-settler
to COD removal. Therefore optimal experiments were
model the parameter that describes the fraction of
designed where extra ammonium was added together with
suspended solids going to the final effluent (fNS) was
the wastewater to simultaneously identify both processes
adjusted to fit the average measured effluent suspended
from one set of experimental data. For a complete
solids concentration (in this case 11 mg SS/l).
description of the lab-scale experiments and their
The experimental value of the decay rate determined
interpretation, including the estimation of kinetic
based on the lab-scale test was 0.41 d − 1 (value corrected
parameters, the reader is referred to Petersen (2000).
for temperature and transformed to the death regenera-
The respirograms were in general not informative
tion concept, as described in (5)). This value was used as a
enough to describe the degradation of COD in the waste-
guideline for the model calibration. The influent for the
water via Monod kinetics, as described in (4) (no zero-
steady state model was obtained by averaging the dynamic
order respiration rate plateau was reached in the
influent data. Initially, these averages were calculated
experiment). Consequently, the degradation of COD was
using only the dry weather flow data (influent data from 19
instead described via a first-order model (17), where the
November 1998 at 10.00 a.m. until the end of the measur-
first-order rate constant k replaces the Monod parameter
ing campaign, and with the wastewater characterisation as
combination mmaxH/KS:
described above). For this period of the measuring
campaign the waste sludge production was calculated to
10,342 kg SS, with an average waste flow rate of 241 m3/d.
Converting into COD units, using the SS to COD conver-
sion factor of 1.38 resulting from the COD analyses done
Step 7–9: steady state model calibration
on the recycle sludge samples, the waste sludge produc-
For the steady state model calibration a simple WWTP tion was calculated to be equivalent to 14,065 kg COD.
configuration was constructed in WEST + + (Figure 13). However, during the initial calibration of the steady state
The steady state configuration consists of an aeration tank model only about half of this waste sludge production
3
(V = 2,600 m ), a point-settler, an internal recycle line could be predicted no matter what the values applied for
 28 Britta Petersen et al. | Evaluation of an ASM1 model calibration procedure Journal of Hydroinformatics | 04.1 | 2002

Table 4 | The average COD composition of the influent during the measuring campaign production of 14,364 kg SS (19,535 kg COD) was

COD component %
obtained, with an average waste flow rate of 248 m3/d. A
new series of steady state simulations now resulted in a
SI 12 waste sludge production of 19,177 kg COD, which was

SS 16
comparable to the waste sludge production measured at
the WWTP (with the wastewater composition of Table 4).
XS 22 The value of the fNS parameter of the point-settler
model was adjusted to 0.23% to describe the measured
XI 50
average SS concentration in the final effluent.
The final calibrated value of bH was 0.5 d − 1 which is
higher than the default value for 10°C (Henze et al. 1987)
but in accordance with the experimental value of
bH and bA were. Adjustment of the specific growth rates 0.41 d − 1. Furthermore, initial adjustments of the
did not solve the sludge balance problem either. The maximum specific growth rates, mmaxA and mmaxH, were
simulated average sludge concentration in the aeration carried out. However, it should be stressed that final
tank and the recycle line were much lower than the values can only be assigned to these parameters in the
measured concentrations. It was tried to increase the dynamic model calibration, which is the last part of the
model waste sludge production by assuming that all model calibration procedure (Figure 1).
particulate COD consisted of XI (instead of 50% as The autotrophic biomass fraction (fBA) in the
initially assumed), but the sludge balance did not improve activated sludge of the full-scale installation can be
sufficiently to solve the problem. approximated by (18) (Sinkjær et al. 1994):
The resulting distribution of the different COD com-
ponents in the influent is summarised in Table 4. Initially,
XI consisted of 25% of the total COD and XS of 47%,
assuming that XS consisted of Srest plus 50% of the
particulate COD, as described above. The calculated fraction became 0.0085 mg CODNIT/mg
We then tried to find other reasons for the imbalance COD. This value is comparable with the steady state
in the waste sludge production. As mentioned above model calibration that yielded an autotrophic biomass
(Figure 5 and 7) the COD and SS concentrations, together fraction of 0.0086 mg CODNIT/mg COD. The fraction of
with the influent flow (Figure 3), were very high during the heterotrophic biomass was calculated similarly and a
first two days of the measuring campaign due to the value of 0.217 mg CODHET/mg COD was obtained, com-
rainfall. This indicates that the primary clarifiers did not parable to a model value of 0.182 mg CODHET/mgCOD.
retain the solids sufficiently during the rain event. This
high load during the rain event could be due to a flush
Step 10: dynamic model calibration
effect in the sewers. This extra load of COD and SS during
the rain event contributed significantly to the sludge The configuration of the dynamic model (see Figure 14)
production, and this extra sludge production was most consists of a plug-flow reactor (6 times 4 reactors in series
probably only wasted during the following dry weather of 100 m3 each), a sludge line (2 reactors in series of
flow days. To take this into account for the sludge balance 100 m3 each), a point-settler, an effluent buffer tank
during the model calibration it was therefore decided to (V = 3,500 m3), and a recycle line (5 reactors in series of
include the data measured during the rain event for the 80 m3 each). The effluent buffer tank was added to the
calculation of the average influent composition. configuration to simulate the liquid retention time in the
The waste sludge production too was recalculated to settlers. However no reactions were assumed to take place
include the data of the rain event. A total waste sludge there.
 29 Britta Petersen et al. | Evaluation of an ASM1 model calibration procedure Journal of Hydroinformatics | 04.1 | 2002

differences) is to compare the value of the above

mentioned parameter combination. With a mmaxA of
0.31 d − 1 and an average simulated XBA of about 40 mg
COD/l, the simulated maximum oxygen uptake rate
becomes 225 mg/l.d. This is in very good agreement
with the parameter combination derived from lab-scale
Figure 14 | Model configuration for the calibrated model.
experiments, with an average of 237 mg/l.d (n = 21) and a
95% confidence interval of 175–300 mg/l.d. The simulated
SNH concentration in the aeration tanks was always
The recycle flow was controlled at 55% of the influent higher than about 3 mg N/l (i.e. the minimum measured
flow rate, to simulate the behaviour of the full-scale concentration of the final effluent, see Figure 9). Thus, in
WWTP. For this purpose an influent flow measurement this case the influence of the half saturation coefficient,
coupled to a proportional underflow controller was KNH, is insignificant.
included in the model. Furthermore, a waste flow control- In the lab-scale experiments the heterotrophic sub-
ler (see Figure 14) was implemented in the model to strate degradation was described with a first-order
simulate the discontinuous sludge waste from the recycle expression since the data were not informative enough to
line. This on/off controller was fed with a data flow that be described with the Monod model, as explained above.
indicated the sludge waste rate (data obtained from plant An evaluation of the calibrated parameters of the full-scale
operation logbook). model versus the lab-scale parameters can still be carried
A problem with the nitrogen balance appeared during out. The substrate degradation rate was calculated based
the rain period, since a rather high effluent SNH peak on both the lab-scale model and its parameters on the one
resulted from the simulations (20–25 mg NH4-N/l). This hand (17), and the Monod model and its parameters on
peak was not observed in the data. To solve the problem, it the other hand (19):
was assumed that the nitrogen content of the high XI load
during the rain period was higher than the assumed 1%.
By adjustment of iNXI to 3% during the rain weather
period the problem of the nitrogen balance could be
solved, because less nitrogen was thus released by The calculated profiles can be seen in Figure 15. The
hydrolysis and ammonification. Note that a fraction of lab-scale parameters are estimated on 16 different lab tests
inert nitrogen of 3% is considerably higher than the typical and the corresponding 95% confidence interval is given in
value of 0.5–1% (Henze et al. 1995). Figure 15 as well. For the Monod model the average
The maximum specific growth rates mmaxH and mmaxA simulated biomass concentration was used for XBH. In the
were calibrated to be 2.8 d − 1 and 0.31 d − 1 respectively lab-scale experiments the identifiable parameter combina-
and the KS was adjusted to 15 mg COD/l. tion was kXBH/YH, and the estimated parameter combi-
The parameter combination involving mmaxA nations were adjusted for the differences in sludge
identifiable from the lab-scale experiments is given by concentrations between the different experiments. It is
clear from (16) that, as SS increases, the model result
based on the lab-scale parameters deviates from the
Monod model. However, it is important to notice that, for
(Petersen et al. 2000), which in fact is equal to the the smaller SS concentrations in the first-order region of
maximum oxygen uptake rate for nitrification, assuming the Monod model, SS < KS, the Monod profile lies within
no substrate limitations. Thus, a way to validate the the results of the lab-scale experiments, confirming that
parameters of the full-scale model with the ones derived values of 2.8 d − 1 for mmaxH and 15 mg/l for KS are reason-
from lab-scale experiments (corrected for temperature able. The simulated SS concentrations in the main part of
 30 Britta Petersen et al. | Evaluation of an ASM1 model calibration procedure Journal of Hydroinformatics | 04.1 | 2002

Figure 15 | Substrate removal rate (dSS/dt) plotted as a function of the substrate

concentration for the Monod model (used in the calibrated model) and the
Figure 17 | Effluent SS data (squares) and model effluent SS predictions.

first-order model (used for the interpretation of respirometric data).

Figure 16 | Effluent SNH data (squares) and model effluent SNH predictions. Figure 18 | Effluent SNO data (squares) and model effluent SNO predictions.

the aeration tanks were indeed below the value of KS. final effluent. Although the activated sludge system is fully
Thus, the experimental first order description of the SS aerated it is likely that some simultaneous denitrification
removal is realistic. can take place in the system, e.g. in the less intensively
Both the description of the effluent SNH and SS data aerated recycle channel. For example, the NO3-N effluent
are good (Figures 16 and 17). Only on one occasion do the concentration decreased to about 1 mg N/l on Friday
simulations result in a SS peak that is not present in the evening and Saturday morning. This was probably due to
data. This SS peak results from a high influent concen- an increased residence time in the aeration tank (lower
tration together with a high flow rate, which could not be flow) combined with the availability of sufficient readily
modelled adequately. One explanation could be that biodegradable carbon for denitrification entering the
degradation of SS could have continued for a while in the WWTP on Friday afternoon (see Figures 3 and 5). The
secondary clarifiers, while the model assumes that no effluent NO3-N concentration increased again on
degradation reactions take place in the clarifiers. Initially Sunday and Monday (Figure 10) due to a lack of readily
the model predicted too high SNO concentrations in the biodegradable COD (see Figure 5).
 31 Britta Petersen et al. | Evaluation of an ASM1 model calibration procedure Journal of Hydroinformatics | 04.1 | 2002

Table 5 | List of the main parameters for the dynamic model (10°C)

Default
Parameter (10°C) Calibrated Units

YH 0.67 g cell COD formed/g COD oxidized

YA 0.24 g cell COD formed/g N oxidized

mmaxH 3.0 2.8 d−1

mmaxA 0.3 0.31 d−1

bH 0.2 0.5 d−1

bA 0.05 0.02 d−1

KS 20 15 g COD/m3

KNH 1.0 g NH4-N/m3

KNO 0.5 g NO3-N/m3

KOH 0.2 0.5 g O2/m3

KOA 0.4 g O2/m3

kh 1.0 g slowly biodegradable COD/g cell COD. d

KX 0.01 g slowly biodegradable COD/g cell COD

hg 0.8 0.6 Dimensionless

The saturation coefficient for oxygen, KOH, was NH4-N (see Table 6). As can be seen from Table 6 the
increased slightly to 0.5 mg O2/l to decrease the inhibition model describes 94–100% of the actual removal, which
of denitrification by O2, and the fraction of denitrifiers, hg, can be considered to be very satisfactory.
was decreased to 0.6, to make the simulated SNO concen-
tration in the effluent approach the measured values.
The resulting description of effluent SNO is not perfect
Sensitivity analysis
but follows the trend of the data reasonably (Figure 18).
Table 5 shows the complete parameter list for the dynamic The parameter values of the calibrated model given in
model. Table 5 did not seem very different from the default
Finally, the removal efficiencies for CODtot, CODsol, parameter set. Therefore, it was investigated if the par-
TKN and NH4-N have been calculated based on the actual ameters that were modified during the model calibration
measured and averaged data on the one hand, and based were indeed influencing the model outputs significantly.
on averaged results of the dynamic simulation including To this purpose a sensitivity analysis was carried out with
the whole measuring campaign on the other hand. From the calibrated model, to check the sensitivity of the model
these results, it has been calculated how close the model output (effluent SS, SNH and SNO concentrations) and the
describes the removal of CODtot, CODsol, TKN and predicted biomass concentration (X) to changes in the
 32 Britta Petersen et al. | Evaluation of an ASM1 model calibration procedure Journal of Hydroinformatics | 04.1 | 2002

Table 6 | Removal efficiency based on measurements and simulation, expressed as The sludge concentration is only significantly influ-
percentage of the observed removal that is described by the model
enced by the value of YH and by the influent XI concen-
Removal efficiency CODtot CODsol TKN NH4-N tration. The latter confirms that a modification of the
influent fractionation (the fraction of XI was increased)
Measurements 84% 73% 68% 58%
was indeed one of the most appropriate things to do to
Simulation 83% 72% 67% 54% increase the sludge concentration in the system (and con-
sequently also the sludge production) during the steady
Model description 99% 100% 99% 94%
state model calibration. Besides the sludge concentration,
YH also influences the output SS, SNO and SNH concen-
trations. However, at the beginning of the calibration it
was decided not to change YH (and YA). The sensitivity
analysis shows furthermore that all but one (hg) of the
model parameters and influent pollutant concentrations.
parameters that were modified from their default values in
The sensitivity was evaluated using relative sensitivity
the final calibrated model (mmaxH, mmaxA, bH, bA, KS, KOH,
functions, RSF (see (20)). The calibrated model was used
hg) influence one or several of the simulated effluent
as the reference simulation, and the model output
concentrations. This confirms (at least for six out of the
obtained after increasing the value of a specific parameter
seven parameters: mmaxH, mmaxA, bH, bA, KS, KOH) that a
with 1% was used to obtain the relative sensitivity
modification of these parameter values resulted in a con-
functions (20):
siderable change in the simulated model output. Finally, it
should be stressed that both mmaxH and mmaxA are very
influential on effluent SS and extremely influential on
effluent SNH and SNO concentrations respectively. In
The value of the relative sensitivity function at the begin- other words, this confirms that even a slight modification
ning of the period with dynamic data was used for the of the value of these parameters, as in the presented model
evaluation ( = output corresponding to the end of the calibration, can have an important effect on the model
steady state simulation). One can comment that this leads output.
to a steady state sensitivity analysis. However, it should be
added here that a similar sensitivity analysis was done for
the period with dynamic data (calculation of the average Model reduction
model deviation ∆y based on the simulation data obtained
It was investigated whether the number of tanks could be
for the period with dynamic data), and that the results of
reduced in the hydraulic model to increase the calculation
this analysis were rather similar to the results of the steady
speed of the model. To evaluate the effect of model reduc-
state analysis presented here. The results of this evaluation
tion, the quality of the fit between simulated values and
are shown in Table 7. The influence of a parameter on the
available data was evaluated by calculating the average
model output was interpreted as proposed by Julien (1997).
relative deviation (ARD) between model predictions of
For RSF < 0.25, a parameter is considered to have no
SNH and available data points (21):
significant influence on a certain model output; if
0.25≤RSF < 1, the parameter is considered to be influ-
ential; if 1≤RSF < 2, the parameter is considered to be very
influential; and if 2≤RSF, the parameter is considered to be
extremely influential. When the value of a parameter and For the calibrated model ARD was 16.5% when consider-
the output change in the same direction, this is indicated ing the effluent SNH concentration. For a reduction of the
with a positive sign in Table 7; when they move in the number of tanks from 24 to 12 in the activated sludge
opposite direction this is indicated with a negative sign. lanes, and from 5 to 3 in the recycle channel, the same
 33 Britta Petersen et al. | Evaluation of an ASM1 model calibration procedure Journal of Hydroinformatics | 04.1 | 2002

Table 7 | Results of sensitivity analysis with the calibrated model of the Zele WWTP. (+, − =influential; + +, − − =very influential; + + +, − − − =extremely
influential). See text for a further explanation of the results

Parameter X Effluent SS Effluent SNO Effluent SNH

YH + − − + − −

YA

mmaxH − −

mmaxA + + + − − −

bH + + +

bA − +

KS + +

KNH − +

KOH −

KOA − +

ka

kh − − +

KX

hg

Influent component

SS − +

SNH + + +

XI +

XS + − − +

ARD (16.5%) was obtained for effluent SNH. Thus, this last model. The reason for this is simply that the mixing
model reduction gave the same accuracy as the full model. patterns are more approaching an ideally mixed situation
However a further reduction of the number of tanks to 8 in compared to the original model. In general, except for the
the activated sludge lanes and 2 in the recycle channel case of zero-order degradation kinetics, an ideally mixed
gave a significantly worse description, with an ARD of tank results in a lower substrate removal efficiency in
73.5%. Although the dynamics of the effluent SNH data comparison with an ideal plug-flow tank when the same
could still be described, the simulated SNH concentrations reactor volume is available. Conclusively, the model
were in general higher than the measured data with the reduction resulted in a model that needed about 50%
 34 Britta Petersen et al. | Evaluation of an ASM1 model calibration procedure Journal of Hydroinformatics | 04.1 | 2002

less calculation time for a simulation than the original reduction study showed that the number of tanks could be
calibrated model. halved, from 24 to 12, but still giving the same description
In principle, one could imagine that, if the value of of the effluent data with the same parameter set as the
mmaxA is increased, then a similar model fit for the reduced calibrated model. However, it also became clear that a
8 tanks-in-series model could be reached as for the further model reduction would not be possible without a
calibrated model, since an increase of mmaxA would compensating change in the kinetic parameters. In other
decrease the outlet SNH(0). Thus, this means that a ‘wrong’ words, for a further model reduction (a decrease of the
hydraulic characterisation could be compensated by a number of tanks below 12) errors in hydraulics have to be
change of parameter values. compensated by ‘wrong’ biological parameters deviating
from the lab-scale results, e.g. increase of mmaxA to
decrease the effluent SNH concentration. Thus, for the
case when a hydraulic model would not have been avail-
able at all, e.g. the hydraulics were described with a
DISCUSSION
4 tanks-in-series model, the calibrated parameter set
In this study it was stressed that the purpose of the model might have been rather different and not corresponding at
should determine how the model is calibrated, e.g. which all to the results of the lab-scale experiments. This is
information is needed and to which level the model should immediately linked to the importance of evaluating the
be calibrated. A systematic and general model calibration key kinetic parameters with lab-scale experiments. In this
procedure was proposed, and concretised for a combined case study the decay rate and the two specific growth rates
municipal–industrial WWTP. ASM1 was applied in the mmaxA and mmaxH were determined. It was illustrated how
case under study, but the proposed general model to compare these parameters obtained from lab-scale
calibration procedure is applicable for any activated experiments with the parameters of the full-scale model,
sludge model. thereby verifying that the parameters of the full-scale
The purpose of the case study was to obtain a good model were realistic.
description of the biological N removal, since the model Thus, as just described above the information
was to be used for process optimisation focusing on an obtained from different tests for hydraulic, sludge settling
improvement of the N-removal capacity, including (if needed) and biological characterisation help to frame
start-up of denitrification (Gernaey et al. 2000). Therefore, the model calibration, and in fact reduces the apparently
it was also important to describe the variation in readily high degree of freedom of the model parameters
biodegradable COD. Biodegradation of COD will influ- significantly.
ence the N components in the activated sludge system, e.g. For this case study, it could be questioned, however,
because SNH is incorporated into new biomass and SNO is whether it was necessary to determine some kinetic
consumed during denitrification. It was observed that the parameters in lab-scale experiments, since the resulting
presence of readily biodegradable COD in the influent was calibrated parameters were not far from the ASM1 default
mainly related to industrial activity, resulting in a lack of parameter set (Henze et al. 1987). Still, even in this case
biodegradable COD during the weekend. For future the lab-scale results gave extra confirmation of the
implementation of denitrification in the WWTP this may parameter set of the calibrated model, thereby increasing
cause problems in maintaining the denitrification effi- the quality and confidence of the model calibration.
ciency during weekends. This weekend effect was reported Moreover, the sensitivity analysis clearly showed that the
previously for another Flemish municipal WWTP (Coen calibrated model was indeed sensitive to changes of the
et al. 1997). parameters that were modified during the model calibra-
In this study it was obviously important to have a good tion procedure. Two of the most influential parameters
description of the hydraulic patterns to describe the were mmaxA and mmaxH, which confirmed that even the
dynamics of the system adequately. Indeed, a model small deviations of these parameters from their default
 35 Britta Petersen et al. | Evaluation of an ASM1 model calibration procedure Journal of Hydroinformatics | 04.1 | 2002

values (Henze et al. 1987) in the calibrated model has a model to describe the plug-flow aeration tank. It was
considerable influence on the model output. It should be shown that this hydraulic model could be reduced to a
stressed that such a sensitivity analysis is case specific, 12 tanks-in-series model, yielding a 50% reduction of the
since the results of the analysis can be influenced by the calculation time for the scenario simulations. Two of the
data set that is studied. This can be illustrated with the most important parameters to adjust to correctly describe
parameter KNH. The calibrated model presented here is the dynamics were the specific growth rates, as was also
not sensitive to a change of KNH because the SNH concen- evidenced by a sensitivity analysis carried out with the
trations in the plant are always considerably higher than calibrated model. Consequently, additional information
the value of KNH (1 mg N/l). However, one could imagine on the specific growth rates derived from lab-scale
that the influence of KNH could be larger for a model that experiments is important to confirm that the calibrated
describes a treatment plant with almost complete nitrifi- parameters of the full-scale model are realistic.
cation (e.g. effluent SNH concentration around 1 mg N/l).
It was clear from this study that there is an interaction
between wastewater characterisation and calibration of
the full-scale model to the available data (effluent and
ACKNOWLEDGEMENTS
sludge wasted). For instance, the influent concentration of
XI and XS were adjusted during the steady state model The authors would like to thank Aquafin NV (Aartselaar,
calibration to be able to describe the sludge production Belgium) for giving us the opportunity to study the WWTP
data. In this phase it became clear that it was important to in Zele, and for providing lab support during the measur-
include the data of the rain weather period prior to the dry ing campaign. Special thanks are due to the technical staff
weather period, since a high COD load originating from of the Zele WWTP for their cooperation at the plant, and
the rain period contributed significantly to the sludge that to Chris Thoeye, Herwig Bogaert and Gunther Parmentier.
was wasted during the following days. Thanks also go to students Annick Beyaert, Andy
Hullebusch and Ingmar Nopens for carrying out parts of
the experimental work during the measuring campaign.
The research of Krist Gernaey was financed with a post-
doc scholarship of the Flemish Institute for the Promotion
CONCLUSIONS
of Scientific-Technological Research in the Industry
A systematic model calibration procedure was presented (IWT, Brussels, Belgium).
and evaluated for a combined municipal–industrial
WWTP. It was underlined that it is very important to
define the purpose of the model carefully since this will
determine how to approach the model calibration. In this
study it was clearly illustrated how additional information NOTATION
obtained from tests specifically designed to describe the ARD Average relative deviation
ASM Activated sludge model
hydraulics and the biology of the system help to decide on
ATU Allylthiourea
realistic model parameters during the model calibration
bA Autotrophic decay rate (1/min)
procedure.
bH Heterotrophic decay rate (1/min)
The aim of this study was to obtain a good description
bH′ Observed endogenous decay rate (1/d)
of the N removal capacity, since the model was to be BOD Biochemical oxygen demand (mg O2/l)
applied for process optimisation at a later stage. It was BOD5 5 day biochemical oxygen demand (mg
thus important to have a good description of the process O2/l)
dynamics. Therefore, the hydraulic behaviour of the sys- BODst Short term biochemical oxygen demand
tem was investigated, resulting in a 24 tanks-in-series (mg O2/l)
 36 Britta Petersen et al. | Evaluation of an ASM1 model calibration procedure Journal of Hydroinformatics | 04.1 | 2002

BODst,total Total short term biochemical oxygen M Mass of Li added (g)

demand (mg O2/l) MLVSS Mixed liquor volatile suspended solids
BODst,NH4 Short term biochemical oxygen demand n Number of experimental data points
for NH4-N oxidation (mg O2/l) N Number of tanks for an N
CNt Normalized Li concentration tanks-in-series model
( = Ct/CO) NH4-N Ammonium nitrogen (mg N/l)
CO Initial Li concentration assuming a NO2-N Nitrite nitrogen (mg N/l)
perfectly mixed tank (g/m3) NO3-N Nitrate nitrogen (mg N/l)
COD Chemical oxygen demand (mg/l) NNIT Amount of nitrified ammonium nitrogen
CODHET Heterotrophic biomass COD (mg (kg N/d)
COD/l) p Value of a parameter for the reference
CODNIT Nitrifying biomass COD (mg COD/l) simulation
CODsol Soluble COD (mg COD/l) PC Primary clarifier
CODsol,effluent Soluble COD in the effluent (mg PO4-P Orthophosphate (mg P/l)
COD/l) rO,end Endogenous oxygen uptake rate (mg
CODsol,influent Soluble COD in the influent (mg O2/l.min)
COD/l) rO,ex Exogenous oxygen uptake rate (mg
CODst Readily biodegradable COD (mg O2/l.min)
COD/l) rS Substrate removal rate (mg COD/l.min)
CODtot Total COD (mg COD/l) RSF Value of relative sensitivity function
CODtot,effluent Total effluent COD (mg COD/l) SALK Alkalinity concentration (meq/l)
Ct Measured Li concentration (g/m3) SBR Sequencing batch reactor
fBA Fraction of autotrophic biomass in the SC Secondary clarifier
activated sludge SI Inert soluble organic material
fp Fraction of biomass converted to XI as concentration (mg COD/l)
a result of decay SI,effluent SI in effluent (mg COD/l)
IE Inhabitant equivalents SI,influent SI in influent (mg COD/l)
iNSI Nitrogen content of SI (g N/g COD) SKN Soluble Kjeldahl nitrogen (mg N/l)
iNXI Nitrogen content of XI (g N/g COD) SND Soluble organic nitrogen concentration
k First-order rate constant (l/mg.min) (mg N/l)
ka Ammonification rate (m3 COD/g d) SNH Ammonium concentration (mg N/l)
kh Maximum specific hydrolysis rate SNI Nitrogen included in inert soluble
(g slowly biodegradable COD/g cell organic material (mg N/l)
COD. d) SNO Nitrite + nitrate nitrogen concentration
KNH Autotrophic half-saturation substrate (mg N/l)
concentration (mg N/l) SO Dissolved oxygen concentration (mg
KNO Half-saturation coefficient for O2/l)
denitrification (mg N/l) Srest CODsol,influent − SI,influent − SS,influent (mg
KOA Autotrophic half-saturation coefficient COD/l)
for oxygen (mg O2/l) SS Suspended solids (mg/l)
KOH Heterotrophic half-saturation SS Readily biodegradable soluble substrate
coefficient for oxygen (mg O2/l) concentration (mg COD/l)
KS Heterotrophic half-saturation substrate SS,effluent SS in effluent (mg COD/l)
concentration (mg COD/l) SS,influent SS in influent (mg COD/l)
KX Half-saturation coefficient for TKN Kjeldahl nitrogen (mg N/l)
hydrolysis of XS (mg COD/l) V Volume of the system (m3)
LiCl Lithium chloride VSS Volatile suspended solids (mg/l)
 37 Britta Petersen et al. | Evaluation of an ASM1 model calibration procedure Journal of Hydroinformatics | 04.1 | 2002

WWTP Wastewater treatment plant Dupont, R. & Sinkjær, O. 1994 Optimisation of wastewater
treatment plants by means of computer models. Wat. Sci.
X Biomass concentration (mg COD/l)
Technol. 30(4), 181–190.
XBA Autotrophic biomass concentration (mg
Ekama, G. A., Dold, P. L. & Marais, G. v. R. 1986 Procedures for
COD/l) determining influent COD fractions and the maximum specific
XBH Heterotrophic biomass concentration growth rate of heterotrophs in activated sludge systems. Wat.
(mg COD/l) Sci. Technol. 18(6), 91–114.
XI Particulate inert organic material (mg Gernaey, K., Petersen, B., Ottoy, J. P. & Vanrolleghem, P. A. 2001
Activated sludge monitoring with combined respirometric—
COD/l)
titrimetric measurements. Wat. Res. 35, 1280–1294.
Xiobs Observed values Gernaey, K., Petersen, B, Parmentier, G., Bogaert, H., Ottoy, J. P. &
Xisim Simulated values Vanrolleghem, P. A. 2000 Application of dynamic models
XND Particulate organic nitrogen (mg N/l) (ASM1) and simulation to minimize renovation costs of a
XNI Nitrogen included in inert particulate municipal activated sludge wastewater treatment plant.
Proceedings of the 1st World Congress of the International
organic material (mg N/l)
Water Association, 3–7 July 2000. AGHTM, Paris, France.
XS Particulate slowly biodegradable
Available on CD-ROM.
organic material (mg COD/l) Gujer, W., Henze, M., Mino, T. & van Loosdrecht, M. C. M. 1999
y Value of model output for the reference Activated sludge model No. 3. Wat. Sci. Technol. 39(1),
parameter set 183–193.
Y Biomass yield Henze, M. 1992 Characterization of wastewater for modelling
of activated sludge processes. Wat. Sci. Technol. 25(6),
YA Autotrophic biomass yield (g COD
1–15.
biomass/g N oxidised) Henze, M., Grady, C. P. L. Jr., Gujer, W., Marais, G. v. R. & Matsuo,
YH Heterotrophic biomass yield (g COD T. 1987 Activated Sludge Model No. 1. IAWPRC Scientific and
biomass/g COD substrate) Technical Reports No. 1. London, UK.
mmaxA Maximum specific growth rate of Henze, M., Gujer, W., Mino, T., Matsuo, T., Wentzel, M. C. M. &
Marais, G. v. R. 1995 Activated Sludge Model No. 2. IAWQ
autotrophic biomass (1/min)
Scientific and Technical Report No. 3, London, UK.
mmaxH Maximum specific growth rate of
Henze, M., Gujer, W., Mino, T., Matsuo, T., Wentzel, M. C., Marais,
heterotrophic biomass (1/min) G. v. R. & van Loosdrecht, M. C. M. 1999 Outline activated
hg Denitrification correction factor sludge model No. 2d. Wat. Sci. Technol. 39(1), 165–182.
∆p Parameter value difference for two Henze, M., Harremoës, P., la Cour Janssen, J. & Arvin, E. 1997
simulations (reference and simulation Biological and Chemical Wastewater Treatment, 2nd edn.
Springer, Berlin.
with modified parameter)
Julien, S. 1997 Modelisation et estimation pour le contrôle d’un
∆y Difference in model output for two procédé boues activées éliminant l’azote des eaux résiduaires
simulations (reference and simulation urbaines. PhD thesis, Institut National Polytechnique de
with modified parameter) Toulouse, France.
OH Hydraulic residence time Kristensen, H. G., la Cour Janssen, J. & Jørgensen, E. 1998 Batch test
procedures as tools for calibration of the activated sludge
Ot Normalized time (t/OH)
model—a pilot scale demonstration. Wat. Sci. Technol. 37(4-5),
OX Sludge age (d) 235–242.
Lesouef, A., Payraudeau, M., Rogalla, F. & Kleiber, B. 1992
Optimizing nitrogen removal reactor configurations by on-site
calibration of the IAWPRC activated sludge model. Wat. Sci.
Technol. 25(6), 105–123.
Ljung, L. 1987 System Identification—Theory for the User.
REFERENCES Prentice-Hall, Englewood Cliffs, NJ.
Coen, F., Vanderhaegen, B., Boonen, I., Vanrolleghem, P. A. & Van Nowak, O., Franz, A., Svardal, K., Muller, V. & Kuhn, V. 1999
Meenen, P. 1997 Improved design and control of industrial and Parameter estimation for activated sludge models with
municipal nutrient removal plants using dynamic models. Wat. help of mass balances. Wat. Sci. Technol. 39(4), 113–120.
Sci. Technol. 35(10), 53–61. Pedersen, J. & Sinkjær, O. 1992 Test of the activated sludge
de la Sota, A., Larrea, L., Novak, L., Grau, P. & Henze, M. 1994 models capabilities as a prognostic tool on a pilot scale
Performance and model calibration of R-D-N processes in pilot wastewater treatment plant. Wat. Sci. Technol. 25(6),
plant. Wat. Sci. Technol. 30(6), 355–364. 185–194.
 38 Britta Petersen et al. | Evaluation of an ASM1 model calibration procedure Journal of Hydroinformatics | 04.1 | 2002

Petersen, B. 2000 Calibration, identifiability and optimal Vanderhasselt, A., Fuchs, A., Vanrolleghem, P., Staudinger, G. &
experimental design of activated sludge models. PhD thesis, Verstraete, W. 1999 Monitoring of the effects of additives on
Ghent University, Faculty for Agricultural and Applied sludge separation properties by using sensors. Wat. Environ.
Biological Sciences, Belgium. Res. 71, 355–362.
Petersen, B., Gernaey, K. & Vanrolleghem, P. A. 2000 Improved Vanrolleghem, P. A. & Coen, F. 1995 Optimal design of
theoretical identifiability of model parameters by combined in-sensor-experiments for on-line modelling of nitrogen
respirometric-titrimetric measurements. A generalisation of removal processes. Wat. Sci. Technol. 31(2), 149–160.
results. In: Ed. Troch, I. and Breitenecker, F. Proceedings of Vanrolleghem, P. A., Kong, Z., Rombouts, G. & Verstraete, W. 1994
the 3rd IMACS Symposium on Mathematical Modelling, An on-line respirographic biosensor for the characterization of
February 2–4, 2000, Vienna University of Technology, Austria. load and toxicity of wastewaters. J. Chem. Technol. Biotechnol.
vol.2, pp 639–642. Argesim, Vienna, Austria. 59, 321–333.
Siegrist, H. & Tschui, M. 1992 Interpretation of experimental data Vanrolleghem, P. A., Spanjers, H., Petersen, B., Ginestet, P. &
with regard to the activated sludge model no. 1 and calibration Takacs, I. 1999 Estimating (combinations of) Activated Sludge
of the model for municipal wastewater treatment plants. Wat. Model No. 1 parameters and components by respirometry.
Sci. Technol. 25(6), 167–183. Wat. Sci. Technol. 39(1), 195–214.
Sinkjær, O., Yndgaard, L., Harremoës, P. & Hansen, J. L. 1994 Weijers, S. R., Kok, J. J., Preisig, H. A., Buunen, A. & Wouda, T. W.
Characterisation of the nitrification process for design M. 1996 Parameter identifiability in the IAWQ model no. 1 for
purposes. Wat. Sci. Technol. 30(4), 47–56. modelling activated sludge plants for enhanced nitrogen
Stokes, L., Takács, I., Watson, B. & Watts, J. B. 1993 Dynamic removal. Computers and Chemical Engineering Suppl.,
modelling of an A.S.P. sewage works—a case study. Wat. Sci. 1455–1460.
Technol. 28(11-12), 151–161. Xu, S. & Hultman, B. 1996 Experiences in wastewater
STOWA 1996 Methoden voor influentkarakterisering—Inventarisatie characterisation and model calibration for the activated sludge
en richtlijnen. STOWA Report 80-96. STOWA, Utrecht, The process. Wat. Sci. Technol. 33(12), 89–98.
Netherlands (in Dutch).

View publication stats