Experience and extensions to the ASM2 family of models

J. Dudley*, G. Buck*, R. Ashley**, and A. Jack***

* WRc, Frankland Road, Swindon SN5 8YF, UK. E-mail: dudley@wrcplc.co.uk
** Department of Civil & Environmental Engineering, University of Bradford, West Yorkshire,
BD7 1DP E-mail: Rashley@bradford.ac.uk
*** United Utilities, Service Delivery, UPM Team, Thirlmere House, Lingley Mere,
Warrington WA5 3LW. E-mail: andrew.jack@uuplc.co.uk

ABSTRACT

The development of ASM2 has created a complex model for biological phosphorus removal. Most of the
published work on calibrating this model has focused on the design of experiments to maximise
information with which to calibrate the model, or the use of hourly data collected around and within an
aeration tank. But many sewage works do not collect such data, nor have such instrumentation. The
application of ASM2 with sparse data collected at a low frequency, and mostly only input-output, is
considered in this paper, based on data collected at Swedish sewage works. This paper shows that ASM2
can be calibrated to reflect the measurements, although some of the parameter values may be outside the
normal range.

This paper also looks at a modification to ASM2d to better handle heterotrophic usage of volatile fatty
acids, and the use of this model to study the effects of large increases in in-sewer storage on sewage
treatment works. Concern about the generation of large quantities of VFAs, and their effect on the sewage
treatment processes, was unfounded.

KEYWORDS

ASM2, activated sludge, modelling, calibration, sparse data, phosphorus removal, VFA, UKWIR, STOAT

INTRODUCTION

The IWA Activated Sludge Model #2 was published in 1994, with formal publication in 1995 (Henze et al., 1995)
and updated in 1998 (Henze et al., 2000). Since its publication there is anecdotal evidence that the model has been
widely accepted, but there have been few papers published giving experience of its use.

Dynamic modelling, which is the area most considered by the users of ASM 2, is commonly seen as requiring
reasonable quantities of accurate, high-frequency measurements - typically, hourly or better data on COD profiles in
an aeration tank, or respiration rate data. Many wastewater treatment plants are not set up to provide either the data
or the measurement frequency. Many sewage works collect data using daily samples, either spot or composite, and
sample the effluent quality and, possibly with reduced frequency, the influent quality. Sampling is usually only
during the period Monday-Friday. The feasibility of calibrating complex models such as ASM 2 to this kind of
sparse data is usually never considered.

The ASM 2 and its main modification, ASM 2d (Henze et al., 2000) allow for heterotrophic bacteria to consume
VFAs or fermentable COD. The preferred consumption of either of these substrates is handled using the standard
switching function approach popularised by the predecessor ASM 1 (Henze et al., 2000). This approach has been
seen as a pragmatic modification to reflect the need for consumption of VFAs and fermentable COD to be affected
by the presence of the other substrate, but, as will be shown, leads to some difficulties with the subsequent
summation of the reaction rates.

This paper thus introduces two unrelated aspects of ASM 2, based on UK experience with using the model: the
application of ASM 2 where data is collected only for routine plant operation, rather than for the specifics of model
calibration, and modifications of ASM 2 to provide a better expression for the heterotrophic usage of VFAs and
fermentable COD. Results from the particular project that led to the ASM 2 model being checked for how COD
competition should be expressed are also given, although this relates more to the issue of hydraulic disturbances and
control of sewage treatment works.
 SPARSE DATA SETS

The first application was to biological phosphorus removal at a Swedish sewage works. This works operates four
activated sludge lanes, and at the time two lanes were using chemical phosphorus removal, with the other two lanes
modified to investigate biological phosphorus removal. The data collected was only for routine monitoring –
composite samples of the influent and effluent, but little additional data. There had been no intent to model the
phosphorus removal system until the STOAT model was introduced as part of a separate project, at which point it
was of interest to see how well the model could be calibrated using the existing historical records.

An existing model was available, based on ASM 1, and the kinetic parameters used for that model, calibrated by the
site staff, were used in the ASM 2 model. The ASM 2 model was calibrated only for the biological phosphorus
removal, and for the solid settling properties.

MODEL CALIBRATION

The model was calibrated by eye, adjusting the parameters until a best fit was judged to have been found. This
procedure has been used successfully in the past. The reasons for this, compared to more formal procedures (such as
least squares assessments) are:
• Where little is known about the underlying error distribution in the data then graphical fits are more robust;
• Outliers are more readily handled through graphical comparison;
• Cut-off data (e.g., ammonia data recorded only as "< 2 mg/l") is better handled through graphical comparisons.
The resulting set of parameters are given in Table 1. Only the italicised items were calibrated for biological
phosphorus removal – the other parameters were calibrated in previous years for previous models used by the plant,
based on ASM 1. Two values are given for the autotroph growth rate as the practice at the works was to use several
values for different parts of the year, as they had a repeatable annual cycle of inhibition caused by some unknown
discharge within the catchment.

Table 1 - Calibration Parameters

Name Value Default
PAO growth rate @ 15°C (1/h) 0.02 0.034
Autotroph temperature coefficient (1/°C) 0.101 0.105
Heterotroph death rate @ 15°C (1/h) 0.015 0.012
Heterotroph death temperature coefficient (1/°C) 0.113 0.069
Autotroph death temperature coefficient (1/°C) 0.113 0.110
Hydrolysis rate @ 15°C (1/h) 0.072 0.102
Hydrolysis temperature coefficient (1/°C) 0.110 0.041
Hydrolysis half-rate constant @ 15°C 0.0007 0.0072
Half-rate constant temperature coefficient (1/°C) 0.101 -0.110
PolyP uptake rate @ 15°C (1/h) 0.09 0.051
O2 half-rate coefficient (autotrophs) (mg O2/l) 0.4 0.5
NH4 half-rate coefficient (autotrophs) (mg N/l) 0.3 1
Maximum P storage (mg P/mg COD) 1.75 0.34
Fractional hydrolysis rate, anoxic conditions (-) 0.4 0.6
Heterotroph yield (mg COD/mg COD) 0.67 0.63
Fraction of inert COD in particulate substrate (mg COD/mg COD) 0.08 0
PolyP requirement for PHA storage (mg P/mg COD) 0.6 0.4
N content of biomass (mg N/mg COD) 0.086 0.07
N content of inert particulate COD (mg N/mg COD) 0.08 0.03

Autotroph growth rate @ 15°C (1/h) 0.0135

Autotroph growth rate @ 15°C (1/h) 0.0208

2
 SIMULATION RESULTS

The results using the calibration data set are given in the following graphs. Data was usually available for the period
Monday-Friday, with weekend data being estimated by linear interpolation between Friday and Monday. Not all
weekdays had data, and missing data was again estimated by linear interpolation between the available
measurements. This adds to the level of uncertainty in the model and explains part of the issue of goodness of fit. It
can readily be seen that for nitrate and ammonia there was good match between the model predictions and the data.
The ammonia results also show the effect of censored data, where there appear to be at least two different laboratory
interpretations: one presenting data as 0.2 mg/l, the other as 0 mg/l, when presumably the measurements indicated
that the result was “less than 0.2 mg/l.”

12
Model
Data
10
Nnitarte (mg/l)itrate (mg/l(

0
01-Jan 31-Jan 02-Mar 02-Apr 02-May 02-Jun 02-Jul 01-Aug 01-Sep 01-Oct 01-Nov 01-Dec 31-Dec

Figure 1 - Effluent nitrate

3
Model
Data
2.5

1.5
Ammonia (mg/l)

0.5

0
01-Jan 31-Jan 02-Mar 02-Apr 02-May 02-Jun 02-Jul 01-Aug 01-Sep 01-Oct 01-Nov 01-Dec 31-Dec

Figure 2 - Effluent ammonia

For phosphorus the modelling results are not quite as good, as can be seen in Figure 3. However it is clear that the
model is representing the basics of the dynamics in the response, indicating when there are periods of high
phosphorus in the effluent. When data was collected for short period at two-hour intervals the results can be seen to
be a much better match, as shown in Figure 4. We assume therefore that accurate modelling of phosphorus is more

3
sensitive to inadequacies in the data set than is the case with nitrogen – but that the model will give reasonable
warning of likely problem periods.

8
Model
7 Data

5
Total phosphorus (mg/l)

0
18-Nov 18-Dec 17-Jan 17-Feb 19-Mar 19-Apr 19-May 18-Jun 19-Jul 18-Aug 18-Sep 18-Oct 17-Nov 18-Dec 17-Jan 17-Feb

Figure 3 - Effluent total phosphorus

1.4

1.2
Model
Data

0.8

0.6
Phosphate (mg/l)

0.4

0.2

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36

Figure 4 - Short-term phosphorus response

4
As is common, MLSS is usually well represented. The greater level of detail given to representing total solids in
ASM 2 has made matching MLSS easier than was the case with ASM 1 – this has been our experience with other
sites as well.

6000

5000

4000

3000
MLSS (mg/l)

2000

Data
1000
Total SS (mg/l)

0
01-Jan 31-Jan 02-Mar 02-Apr 02-May 02-Jun 02-Jul 01-Aug 01-Sep 01-Oct 01-Nov 01-Dec 31-Dec

Figure 5 - MLSS

SENSITIVITY ANALYSIS

The results of the previous sections have shown that it is possible to calibrate ASM 2 using relatively sparse, routine
data and get reasonable results from the model. The various parameter values used are not outside the normal range
of experience, with the exception of the maximum phosphorus storage – 1.75 mg P/mg COD, compared to a default
value given in the ASM 2 report of 0.34 mg/mg. This value is not only much larger than the default but appears to
be stretching the bounds of physical reality. The relevant equation using this parameter in ASM 2 is:

K max − X PP / X PAO
q max ⋅ ⋅ (!)
K IPP + K max − X PP / X PAO

The results of a sensitivity analysis on the maximum phosphorus storage, Kmax, are given in Figure 6. This shows
that at the default values, and at values of Kmax < 1, the model predictions are unacceptable compared to the data.
Only as Kmax exceeds 1 do the model predictions begin to approach the measured data. During this simulation period
the concentration of PAOs was predicted to be in decline (see Figure 7) – the high value for Kmax, however, kept the
concentration of polyphosphate more or less constant across the simulation period. Thus, the high value for Kmax is
needed to compensate for a modelled loss of biomass. There is nothing in the available data that readily explains the
sudden fall-off in the PAO concentration. This fall-off in concentration does not explain the high values of
phosphorus being predicted at the start of the simulation period, when PAO levels were high, when the default value
for Kmax was used. Thus we are left that the high value for Kmax appears to be fulfilling the functions of an atypically
high value required to ensure that the model represents the data, and a value created as an artefact to compensate
either for deficiencies in the model structure or the available data (whether in terms of operational practice, influent
flows and quality, or whatever.)

5
 10

Default
9 Kmax = 0.7
Kmax = 1
Kmax = 1.5
8
Kmax = 1.6
Kmax = 1.7
7 Kmax = 1.9
Data

5
Phosphate (mg/l)

0
0 200 400 600 800 1000 1200 1400
Elapsed time (h)

Figure 6 - Sensitivity Analysis

Figure 7 - Variation of PAOs in time and space

6
 HETEROTROPHIC USAGE OF VFAS

ASM 2 (Henze et al., 1995) has two equations to describe the usage of VFAs and fermentable COD by heterotrophs.
The essential aspects of these equations are that the choice of the usage of VFA or fermentable substrate is made
using the following switching functions

SA SA
Usage of VFA: ⋅
K A + S A S A + SF

SF SF
Usage of fermentable substrate: ⋅
KF + SF S A + SF

The default values for the switching constants KF and KA are KF = KA = 4 mg/l COD. With these default values
there is no differentiation between the consumption of VFAs and fermentable COD, and the resulting soluble COD
S − SF S − SF SF S
consumption equation has the form ⋅ + ⋅ F . The standard form should be
K A + S − SF S KF + SF S
S
. Assuming SF = SA then the ASM2d switching function has the correct values at the limits S>>K but
K+S
S
underpredicts for S << K, with the equation simplifying to .
2K + S
A better form for the switching function has been known in the standard biochemical engineering literature (Bailey
and Ollis, 1986)1 and has the form

SA / KA
Usage of VFA
1+ SA / K A + SF / KF

SF / KF
Usage of fermentable substrate
1+ S A / K A + SF / KF

S
For the special case KA = KF this equation reduces to , as required. For general modelling the default values
K+S
for the switching constants were made KA = 4 mg/l and KF = 20 mg/l. The value for KF is in line with the value used
in the previous ASM 1 and is a value that has commonly been quoted for heterotrophic usage of soluble COD.
Penya-roja et al. (2001) also concluded that a better value for KF was around 20 mg/l, based on respirometric
calibration of ASM2d on Spanish wastewaters.

IN-SEWER STORAGE SYSTEMS

The issue of the correct representation of how heterotrophs should use VFAs and fermentable substrate arose from
concerns over the adoption of large in-sewer storage to minimise sewer overflows. With large storage volumes,
potentially stored for long periods, there was a possibility of anaerobic conditions leading to the formation of high

1
One other popular phosphorus removal model, that of Barker and Dold (2000), uses a different set of
switching functions to distinguish between heterotrophic usage of VFAs and fermentable COD. The
switching functions used in the Dold model also fail to produce the correct equation form for the special case
KA = KF, and the Dold model should similarly be modified to use the Bailey and Ollis expression. In the case
of the Dold model the switching function reduces to 2 S / (2 K + S) with the same assumption of SA = SF. At
low S this simplifies to the correct for, S/K (compared to ASM 2 of S/ 2 K), while at high S it simplifies to 2,
compared to ASM 2 (and correct), 1.
concentrations of VFAs. When these VFAs were discharged and reached the sewage works they may cause odour
problems and, being perceived as more degradable than other COD forms, may also lead to extensive near-anaerobic
conditions near the start of aeration lanes or the inlet part of filters. This concern led to an UK Water Industry
Research-funded project to examine the effects of large amounts of in-sewer storage on sewage works (Ashley et al.,
2001).

It was necessary to model long term rainfall events to study prolonged loading and dilution effects and also shorter,
more intense storms with foul flushes. Selection of long-term model rainfall inputs was based on the method
proposed by Jack and Ashley (2001), known as the ‘Total Emission Analysis Period’ (TEAP). A critical sequence of
wet and dry weather was selected based on historical annual data (drawn from a 10-year series). An existing
HydroWorks sewer flow model was used for a verified system for which storage has recently been introduced. This
model was adjusted to simplify the level of detail in the sewer network, and to investigate the effects of different
storage volumes and changing the location of the storage position within the sewer. Sewer flow quality was
modelled using a combination of HydroWorks (solids and nitrogen) and the Danish WATS model (carbon and
oxygen transformations) (Hvitved-Jacobsen et al., 1998). STOATTM was used to model the sewage treatment works.

The WATS modelling discovered that unless storage was close to the sewage works the VFAs were rapidly oxidised
during transport through the sewer, and were never present at concentrations high enough to be an odour or other
problem2. When the storage was considered as being almost next to the sewage works then larger quantities of VFAs
(and hydrogen sulphide) were present at the sewage works inlet – but still at concentrations low enough that they
were not an issue. As an example, the summary results for a comparison between running with VFAs present in the
influent, and with the VFAs subsumed into the fermentable substrate, are given in Table 2

Table 2 - Effluent quality impact of VFAs in the influent

With VFAs Without VFAs
TSS COD NH3 TSS COD NH3
Mean 12.8 25.3 1.7 12.8 24.4 1.0
Minimum 4.0 10.4 0.1 4.0 10.4 0.1
Maximum 79.4 80.7 7.9 80.7 81.9 6.9
Standard deviation 11.2 11.6 1.9 11.2 10.2 1.2
It can easily be seen that there is no effect on suspended solids - hydraulic effects on the final settling tank affect this
more, so unless the VFAs produce a large change in the MLSS the effects will be negligible. For COD there has
been a small change, but again the effect is minimal. With ammonia there is a larger difference - the mean value is
almost doubled, caused by the greater affinity for VFA over fermentable substrate, so that at one critical point the
aeration system operated for a longer period with lower DO than the no-VFA case. Had the modelling used higher
values for the KLa, or experimented with the controller settings to provide a more responsive control system, then
this affect may have been less marked.

The rest of the modelling for this project focused on the use of control strategies to reduce the hydraulic effects of
the prolonged periods of storm flow caused by the sewer storage. The aeration system used was set up to be
approximately 8 hours retention at average flow, 8 day sludge age, 4 stages of which the first was 10% of the
volume and anoxic and the other three all aerated and 30% of the volume. The results can be summarised as follows:
• Operating with flow-proportional RAS generally produced good results. For one set of operating conditions and
storm flows, however, this produced a worse effluent quality than operating at a fixed RAS flowrate.

2
The analysis assumed that the storage was provided in-line, that is, as part of the main sewer flow. Off-line
storage, where the flow is diverted to a separate holding tank, may potentially have sewage standing for
longer, producing more VFAs, and may adjust slightly the conclusions reached here. There would still be a
rapid consumption of VFAs and hydrogen sulphide after discharge, provided that some part of the sewer
was aerobic, so that the general conclusion – that VFAs produced through anaerobic storage conditions are
usually ‘treated’ by the sewer system itself – is probably still valid.

8
• Step feed was a viable solution and did not affect the ammonia values much3, probably because of the dilution
of the ammonia in the influent during the storm periods. The results are similar to those of Chen and Beck
(2001). They found that step-feed produced better effluent quality than flow-proportional RAS when looking at
TSS and COD, but worse when looking at ammonia. The comparative results from our modelling exercise are
given in Table 3
• Running at a higher DO setpoint did not have much effect
• Bypassing the primary tanks during storms, or having a storm overflow after the primary tanks (so that all storm
flows went through the primary tanks) did not have much effect.
• Having a sacrificial lane to handle a large share of the storm flow did not produce any significant improvement
in either effluent quality or recovery time following the end of the storm.
• Modelling single storm events did not stress the modelled works as much as having a sequence of storms.
Although the effluent quality indicated a quick return to normal operation after the first storm event, there was
still disturbance of the internal states. Because of this, a second storm event following in quick succession
produced a larger effect on effluent quality than the first. The storm events being used were taken from a
particularly wet period for the catchment area being used.
• A trickling filter model was also used. Although this did not produce as good an effluent quality as the
activated sludge systems (something which can be addressed by changing the filter bed area, depth or packing
characteristics) there was very little effect of the different storage scenarios on effluent quality, unlike the
activated sludge system. The filter system appears to be robust to hydraulic upsets, more so than activated
sludge systems.

Table 3 - Performance with variable RAS and step-feed

Variable RAS Step feed
TSS COD NH3 TSS COD NH3
Mean 16.4 30.4 0.8 10.6 24.8 1.2
Minimum 3.0 10.6 0.1 3.3 9.5 0.1
Maximum 169.3 179.2 7.8 69.5 64.3 9.2
Standard deviation 17.6 16.5 1.1 8.1 10.3 1.4

CONCLUSIONS

Complex models such as the ASM 2 can be calibrated against data that has been collected for routine site monitoring
rather than to assist model calibration. Such data is typically low frequency – typically, daily – with many holes –
typically, week-days only and even then with missing days. Consequently the driving function for the model – the
influent flow and quality – has to make use of much interpolated data. The goodness of fit between the model
predictions and the measured effluent quality will usually be reduced because of this, regardless of the quality of the
calibration of the adequacy of the model representations.

As used at one site, we found that it was possible to get a good match on nitrogen and a reasonable match on
phosphorus. The phosphorus results are possibly better used qualitatively, to indicate likely periods of with- and
without-consent effluent. The parameter values for the calibrated model are usually close to the recommended
default, or within a range that leads to little comment about the values, other than for the maximum phosphorus
storage. This, at 1.75 mg P/mg COD, appears to be too high for comfort. However, a sensitivity analysis on this
parameter indicated that the calibrated value was required to acceptably represent the dynamics of the process, while
basic analyses of the influent data and operating conditions did not show up any explanation for why this parameter
was required to be so large.

The standard representation of the equations for heterotroph utilisation of VFAs and fermentable COD is incorrect,
and fails to meet simple analytical requirements. A simple, well-known, correction to the equations produces the
correct theoretical behaviour while adding no additional complexity to the ASM 2 family of models.

3
Step feed reduces the aeration volume available for nitrification, and this reduction can lead to reduced
nitrification.

9
Although there is this theoretical weakness, in practice ASM2 is normally used with biological phosphorus removal
systems, where most of the VFAs would be consumed in the anaerobic zone. Consequently, for practical purposes
this theoretical weakness of the ASM 2 is unlikely to require that any previous modelling work would need to be
revisited.

Of several control strategies tried to improve performance under storm conditions only two, step feed and variable
RAS flows, were found to provide any useful improvements in effluent quality.

REFERENCES

Ashley, RM, J Dudley, J Vollertsen, A Saul, A Jack, and JR Blanksby, 2001, "The effect of extended in-sewer
storage on wastewater treatment plant performance", Water Science and Technology (accepted); Proceedings of the
2nd International Conference on Interactions between Sewers, Treatment Plants and Receiving waters
(INTERURBA II)

Bailey, JE, and DF Ollis, 1986, Biochemical Engineering Fundamentals, 2nd edition, McGraw Hill,
ISBN 0 07 Y66691 6

Barker, PS, and PL Dold, 1997, "General model for biological nutrient removal activated sludge systems: model
presentation", Water Environment Research 69(5) 969ff

Chen, J and MB Beck, 2001, 'Operational control of storm sewage at an activated sludge process', Water Science
and Technology 43(7) 131-137

Henze, M, W Gujer, T Mino, T Matsuo, M Wentzel and GvR Marais, 1995, Activated Sludge Model No. 2, IAWQ
Scientific and technical report No. 3, ISBN 1 900222 00 0

Henze, M, W Gujer, T Mino and M Loosdrecht, 2000, Activated Sludge Models ASM1, ASM2, ASM2d and ASM3,
IWA Scientific and Technical Report No. 9, ISBN 1 900 222 24 8

Hvitved-Jaobsen, T, J Vollertsen and N Tanaka, 1998, 'Wastewater quality changes during transport in sewers - an
integrated aerobic and anaerobic model concept for carbon and sulphur microbial transformations", Water Science
and Technology 38(10) 257-264

Jack, A, and RM Ashley, 2001, "The impact of the controlled emptying of in-sewer storage on wastewater treatment
plant performance", Water Science and Technology (accepted); Proceedings of the 2nd International Conference on
Interactions between Sewers, Treatment Plants and Receiving waters (INTERURBA II)

Penya-roja, JM, A Seco, J Ferrer and J Serralta, 2001, 'Calibration and validation of activated sludge model No. 2d
for Spanish municipal wastewater", submitted to Environmental technology Letters

ACKNOWLEDGEMENTS

The authors would like to thank the UK Water Industry Research for funding part of the work reported in this
project, and the European Union for funding for other work reported here. JD and GB would also like to thank WRc
plc for permission to publish the material in this paper. The views expressed here are not necessarily those of
UKWIR, the EU, WRc plc, Bradford University or United Utilities.

10