L. Benedetti*, D. Bixio** and P.A.

Vanrolleghem*
*BIOMATH, Ghent University, Coupure Links 653, B-9000 Ghent, Belgium
(E-mail: [email protected])
**Aquafin nv, Technology Department, Dijkstraat 8, B-2630 Aartselaar, Belgium
Abstract Numerical models can be used to evaluate design and upgrade scenarios of urban wastewater
systems on the basis of their ecological consequences. The objective of this paper is to illustrate a
systematic procedure of system design/upgrade. This procedure consists of the following steps: (1) data
collection and data reconstruction; (2) model building and calibration; (3) evaluation of scenarios; (4)
uncertainty assessment. In contrast to conventional practice, this approach allows to choose the most
appropriate trade-off between cost of measures and risk of non-compliance with regulatory limits. An
example of its application dealing with the assessment of WWTP design and upgrade options is provided.
Results show that by reducing the tank volumes compared to conventional design procedures, costs can be
reduced sensibly while the risk of not meeting legislative requirements are only slightly increased.
Keywords Cost-effectiveness; mathematical modelling; Monte Carlo simulation; risk-based design; urban
wastewater systems

Introduction

The European Water Framework Directive (WFD) requests the achievement good quality
ground and surface waters by organising water management on a river-basin scale and,
with regard to impacts on natural water bodies originating from wastewater release,
applying a combined emission and water quality based approach. There is therefore a
need to evaluate and quantify what potential costs and benefits could result from the
WFD approach, which is setting water quality goals in the natural water bodies instead of
prescribing the design of urban wastewater systems. With this new water quality based
approach, the design of the systems is by far less predetermined and the options to meet
the goals become much more numerous. Therefore, new design methodologies must be
developed in order to be able to cope with such increased complexity. The EU project
CD4WC (www.cd4wc.org) is currently tackling these problems, focussing on the urban
wastewater system.
This paper outlines a new methodology to identify and quantify the costs and benefits
for the development of the urban wastewater system resulting from the WFD approach,
with regard to its ecological and economic consequences. Criteria to assess the ecological
consequences are, besides the water quality, also secondary resource inputs, such as
energy, materials and chemicals.
The developed methodology is illustrated in this paper by a case of WWTP design, with
comparisons between three process options and between different activated sludge tanks
volume for one process option. Traditionally, treatment plants have been designed using
empirical steady-state equations or rules of thumb, introducing conservative safety factors. Such an approach has led to the construction of overdimensioned, expensive, and not
always properly functioning, plants. To design or upgrade a treatment plant, deterministic
models can be used since their parameters have a straightforward physical meaning and
doi: 10.2166/wst.2006.571

Water Science & Technology Vol 54 No 67 pp 371378 Q IWA Publishing 2006

Assessment of WWTP design and upgrade options:

balancing costs and risks of standards exceedance

371

can be directly measured in the system or applied to it (in the case of obtained volumes,
recycle rates, etc.).
The software tool used in this study is able to deal with complex models including
draining catchment, sewer system, WWTP and receiving water (Benedetti et al., 2004;
Solvi et al., 2005; Vanrolleghem et al., 2005).
L. Benedetti et al.

Methods

The methodology consists of four steps: (1) data collection and reconstruction; (2) model
building and calibration; (3) evaluation of alternatives; and (4) uncertainty assessment.
Data collection and reconstruction

Data collection is probably the most challenging task. In order to evaluate upgrade
alternatives, long term (several representative years) dynamic influent data are advisable.
In absence of these data, influent data can be reconstructed using available measurements
and making assumptions on the influent properties (Bixio et al., 2002). In the proposed
procedure, a dynamic influent is generated using a simple model of the draining catchment (Gernaey et al., 2005), which considers the number of inhabitants, the presence of
industry, the loads per capita of households and industry, the size of the catchment, the
length of the sewer system, rainfall data and the interactions with groundwater.
Model building and calibration

The most used models for biological treatment are the IWA activated sludge models
(ASM) (Henze et al., 2000). Other important parts of the plant that need to be modelled
are secondary settlers, aeration systems and, when present, anaerobic digesters.
In the case of a renovation or upgrade, relevant information to build and calibrate
these models is often available. For such situations, calibration methodologies can be
found in a large number of publications, for the most comprehensive see Hulsbeek et al.
(2002), Langergraber et al. (2004), Melcer et al. (2003), Vanrolleghem et al. (2003). This
calibration should be performed by experienced modellers to limit the possibility of misinterpretation of the model results. After having obtained a calibrated model of the plant,
several upgrade options can be implemented by modifying and/or extending the plant
model configuration, by increasing or adding volumes, modifying flow paths, adding or
changing processes, etc.
In the case of a design (i.e. no infrastructure is in place yet), standard parameter values
should be applied for the used models, while operational variables are decided by the
modeller, and optimal values for them will be identified by performing several simulations with different values, or applying an optimisation algorithm. In this case, there is
no calibration of the model to be performed, but only some adjustments to parameter
values; for example if for any reason (temperature, influent characteristics, etc.) the
expected Sludge Volume Index is higher than average, some parameters in the secondary
settling model should be modified.
Evaluation of scenarios

The comparison of alternative scenarios can be based on performance criteria that can be
grouped into two categories: environmental and economic criteria. The weight attributed
to them in the decision making process depends on the specific situation of the project.

372

Environmental performance. The proposed methodology adopts the approach set out by
IWA (Copp et al., 2002). It consists of the evaluation of three indicators: the effluent
quality index, the time and number of effluent violations and operating costs.

L. Benedetti et al.

The effluent quality index (EQI) is meant to quantify the effluent pollution load to a
receiving water body in a single variable. The EQI is the weighted sum of the pollution
loads due to: (1) total suspended solids (TSS); (2) chemical oxygen demand (COD); (3)
biological oxygen demand after 5 d (BOD5); (4) total Kjeldahl nitrogen (TKN); (5)
nitrates; and (6) total phosphorus (TP) over one complete year. The used weights are
based on Vanrolleghem et al. (1996) who cited a Flanders effluent quality formula for
calculating fees. Note that the formula is not based on actual environmental damage but
on the operating cost to treat the pollutants.
Effluent violations are calculated for five terms: ammonia, total nitrogen (TN), BOD5,
COD and TSS. The number of violations, as well as the percentage of time that the constraints are not met, is calculated from the output data generated at 15 min intervals.
Operational issues are considered through four items: sludge production, pumping
energy, aeration energy and mixing energy. Sludge production is accounted as mass of
TSS to be disposed, while energy is in kWh consumed.
Economic performance. The evaluation of costs for wastewater treatment is very
complex. In a European context, costs can differ among countries or regions because of
different specific conditions and also because of differences in planning and building
procedures (Bode and Lemmel, 2001). This complexity makes the approach to calculate
costs in order to compare different plant configurations and operational strategies very
difficult. Detailed cost calculations should, in general, be preferred over the use of cost
functions, which can only be useful for rough estimations. Most WWTPs are tailored to
specific conditions/needs, i.e. plants with the same treatment performances do not
inevitably incur the same costs. The use of cost functions is feasible only for process
options screening (Gillot et al., 1999), as it is the case here. In this project, operating
costs have been estimated with the benchmark assessment procedure (Copp et al., 2002)
and with prices representative for northern Europe, while the capital costs were provided
by Ruhrverband (Germany).
Uncertainty assessment

It is an increasingly common practice to use deterministic dynamic models to evaluate

design and renovation scenarios of WWTPs. One of the remaining issues when dealing
with these deterministic models is the degree of uncertainty linked to their predictions.
Probabilistic design, which is the combination of probabilistic modelling techniques
with the currently available deterministic models, provides a solution to this issue (Bixio
et al., 2002). By building a probabilistic shell around the deterministic models, one can
quantify the uncertainty of the model predictions. For example, a goal can be to determine
the probability of exceeding the legal effluent standards of a WWTP. This percentage of
exceedance should be accompanied by confidence intervals indicating the uncertainty due
to the variability of influent characteristics and to the uncertainty in model parameters.
The quantification of the uncertainty of the system as a whole may be carried out by
the following steps (Rousseau et al., 2001): (1) assigning information about the probability distribution of each parameter and input variable in the system; (2) for every simulation with the deterministic value, the simulator uses a value for each parameter and
input variable (a shot) that is randomly selected by a Monte Carlo engine from the
appropriate probability density function. Over multiple simulations, the Monte Carlo
engine produces a range of values for the parameters and input variables that cover the
probability density function; (3) the deterministic model is solved for each shot, as it
would be for any deterministic analysis; all these simulations are independent from each
other, and can thus be run in parallel.

373

L. Benedetti et al.

This iterative process generates a probability density function or cumulative density

function of the output. Based on the distribution of the output, a risk level representing
the high end (e.g. 95th percentile), central tendency (median or mean), or any other
desired level of probability can be identified. See Figure 1 for two examples of uncertainty visualisation.
The number of necessary simulations for scenario analysis tends to be large, especially
when Monte Carlo-based uncertainty assessment is also considered. Hence, a major limitation for a more widespread use of probabilistic design techniques is the calculation time
required. To reduce this computational burden, tools that distribute simulations over
idling PCs available in a local network are under development (Claeys et al., 2005).

Results and discussion

The procedure is illustrated for the case of WWTP design, either for the comparison
between process options or between different activated sludge tank volumes for one process option.
The chosen catchment characteristics (generating the wastewater to be treated) refer to
a combined sewer system collecting 30,000PE (of which 15% of COD load is industry),
located in an area with oceanic climate conditions (characterised by an actual rainfall series
for a representative year with 10 min interval, and a daily temperature series), with infiltration as 30% of dry weather flow, having a weekend load increase of 12% and a summer
load decrease of 25%. All loads are taken from ATV (2000). The TN load is increased by
10% above the ATV load to consider the presence of reject water from the sludge treatment line. The influent time series was generated with the model of Gernaey et al. (2005).
The used rainfall series and an example of WWTP influent are shown in Figure 2.
Three process configurations are compared: low loaded activated sludge (LLAS),
LLAS with primary settler (PS) and oxidation ditch (OD), all equipped with simultaneous
chemical phosphorus precipitation. Such processes have been dimensioned according to
the guidelines contained in ATV (2000), in order to represent existing treatment plants,
and modelled using ASM2d (Henze et al., 2000).
In this work, the number of parameters considered as uncertain was 13 (11 belonging
to ASM2d and two to the influent fractionator model), and for each configuration 60
Monte Carlo simulations were run; this number was sufficient to reach the convergence
110

110

100

100
50%

90
AEC [ /d]

AEC [ /d]

90
80
70
60

50%

70
60

75%

50

75%

50

40

40

100%

30

100%

30
0

374

80

3
4
NH4 [mg/L]

3
4
NH4 [mg/L]

Figure 1 Two options to visualise Monte Carlo simulation results: all results as a cloud of markers (left) and
polygons joining the 5th and 95th percentiles for the two variables and the 50th percentile as a marker
(right); the data show effluent NH4 and aeration costs for different tank volumes (as percentage of
conventional dimensioning volume)

TKN [mg/L]

80
70
60
50
40
30
20
10
0

L. Benedetti et al.

Rain [mm/10 min]

6
5
4
3
2
1
0
185

205

225

245

265

285

305

325

345

365

Time [d]

Figure 2 Rainfall for oceanic climate (down) and TKN influent of the 30,000PE plant (up)

of the output frequency distributions. The natural variability of the influent is assumed to
be represented in the 1 year time series with 15 min intervals.

Comparison of different process alternatives

The performance of the three simulated processes is summarised by the EQI and by the
total cost (sum of aeration, pumping, mixing, chemicals, sludge treatment and disposal
and annualised capital costs). In Figure 3, such indices for the three processes are shown
for the dimensioning according to the ATV guidelines (left) and for a volume reduction
to 70% of the biological treatment tanks (right).
From Figure 3, several conclusions can be drawn. First of all, apparently LLAS with
PS is not a suitable configuration in this situation, since it shows higher costs without giving additional removal performance; note that this conclusion no longer holds when
anaerobic digestion is also present, to which primary sludge is of great benefit. The OD
performs slightly better then LLAS, both in economic and environmental terms, but it
shows a larger uncertainty in its behaviour, and this is probably due to the fact that in the
used model configurations, only very basic control features were implemented (dissolved
oxygen set-point for all of them, with fixed aerated phase length for the OD), while the
OD process performs more stably with additional control systems (e.g. variable aerated
phase length based on NH4 measurements). Furthermore, a reduction of volumes seems
1350

1350

1300

1300
LLAS PS

1250

1250
1200

1200

1150
LLAS

1100
1050

TC [ /d]

TC [ /d]

1150
OD simP

1100

1000

950

950

900

900

850

850
800
1.6

800
1.6

1.7

1.8

1.9

2
2.1
EQI [ton/d]

2.2

2.3

2.4

2.5

LLAS PS

1050

1000

LLAS 30
OD simP

1.7

1.8

1.9

2
2.1
EQI [ton/d]

2.2

2.3

2.4

2.5

Figure 3 Comparison of three alternatives: volumes from ATV guidelines (left) and 70% of biological tanks
volume (right)

375

1100

0.5
100%

100%

1050

0.45
90%

90%

950

CH [ton/d]

TC [ /d]

1000

80%

900

70%

80%

0.4

70%
60%

0.35

L. Benedetti et al.

50%

60%

850

0.3
800
50%
750

2.1

2.2

2.3

2.4

2.5

2.6

2.7

0.25
1.64 1.66 1.68

1.7

EQI [ton/d]

1.72 1.74 1.76 1.78

1.8

1.82

TP [mg/L]

Figure 4 Comparison of different volumes for LLAS: EQI and total cost (left), TP and P-precipitant
dosage (right)

to be more beneficial to a reduction of costs than it is detrimental to the removal of

pollutants.

Comparison of different tank volumes

100

100

90

90

80

80

70

70

Exceedance [%]

Exceedance [%]

The LLAS process was further investigated by simulating it with several activated sludge
tank volumes, ranging from 100 to 50% of the volume calculated with the ATV guidelines,
with 5% intervals for a total of 11 models to simulate. In Figure 4, two comparisons are
presented. It can be deduced that the removal of pollutants only deteriorates significantly
from a volume smaller than 60% of the original one. Of course, the total costs are constantly reduced. In particular, with smaller volumes the TP removal improves, also leading
to a reduction of chemical dosage. Figure 1 shows that, on the other hand, smaller volumes
produce higher ammonia effluents and require higher aeration costs.
Another use of the proposed methodology can be the comparison of concentrationexceedance curves, as illustrated in Figure 5 and Table 1. Figure 5 compares the concentration-exceedance curves calculated with 24 h (left) and 2 h (right) averages of
TN concentrations for the LLAS plant with 70% of the ATV volume. It shows that less
effluent variability is revealed with 24 h averages (curves are more vertical), which leads in
this case to a conservative estimate of the risk of exceedance of, for example, a limit
of 15 mgTN/L, compared to the use of 2 h composite samples as referred to in the German
legislation. This is more evident in Table 1, where for the same plant the risk of exceedance of 15 mgTN/L is given as 5th, 50th and 95th percentiles; direct evaluation of the

60
50
40
30

50
40
30

20

20

10

10

10

12

14

16

TN [mg/L]

376

60

18

20

22

24

10

12

14

16

18

20

22

24

TN [mg/L]

Figure 5 Comparison of TN concentration-exceedance curves for LLAS with 75% of ATV volume: 24 h
(left) and 2 h (right) averages; full line: 50th percentile, dotted lines: 5th and 95th percentiles

Table 1 Exceedance risk as percentage of time of 15 mgTN/L for LLAS with 75% of ATV volume

15 min data
2 h average
24 h average

5th percentile

50th percentile

95th percentile

2.3
2.2
0.4

7.4
7.3
3.7

17.0
16.9
11.4

L. Benedetti et al.

Increasing
volume

Increasing
volume

Figure 6 Comparison of different volumes for LLAS: percentage of time with NH4 emission exceedance
(95th percentile) confronted with operational costs (left) and with total costs (right) and exceedance period
as percentage of the total period

original 15- min interval data (which can be considered as data coming from an on-line
monitoring system) is also included. It can therefore be argued that a risk-based dimensioning is dependent on the local legislation, which can require different criteria to be fulfilled
(yearly average, 2 h average percentiles, daily maxima, etc.).
To resume the economic evaluation, the surfaces in Figure 6 represent, for different
daily costs and different NH4 concentrations, the 95th percentile of the exceedance period
(as a percentage of the total period of 1 year). For example, it can be deduced that with the
ATV volume (corresponding to an operating cost of 380e/d), one can be 95% sure that the
NH4 concentration will not exceed 2 mg/L for more than 18% of the time, while with an
operating cost of 500e/d (corresponding to 50% of the ATV volume), one can be 95% sure
that NH4 concentration will not exceed 2 mg/L for more than 73% of the time. Hence,
smaller volumes lead to substantially lower total costs, although the small disadvantages of
increased NH4 emission and lower process stability have to be taken into account.
Conclusions

A methodology for the assessment of urban wastewater systems has been outlined. The
particular case of a WWTP design served as an example to illustrate the procedure. The
capabilities of the software tools used allow the investigation of the cost-effectiveness of
measures for WFD compliance by using a dynamic integrated model including draining
catchment, sewer system, treatment plant and receiving water. As for the WWTP dimensioning analysis, it can be concluded that building activated sludge volumes smaller than
the ones traditionally calculated by applying design guidelines leads to considerable total
cost reductions, while it entails only a small increase in the risk of not complying with
emission standards. The same considerations can be applied to cases of plant upgrade,
where the increase of risk can be evaluated for changes in the plant loading. The actual
availability of well-accepted models, risk assessment techniques and sufficient computational power should move the design practice from the use of conservative and expensive rules to more advanced and cost-effective procedures.

377

Acknowledgements

The results presented in this publication have been elaborated in the frame of the EU projects CD4WC, contract no EVK1-CT-2002-00118 and Harmoni-CA, contract no EVK1CT-2002-20003.

L. Benedetti et al.
378

References
ATV (2000). Standard ATV-DVWK-A 131E. Dimensioning of Single-Stage Activated Sludge Plants.
Benedetti, L., Meirlaen, J., Sforzi, F., Facchi, A., Gandolfi, C. and Vanrolleghem, P.A. (2004). Dynamic
integrated modelling: a case study on the river Lambro. In: Proceedings of NOVATECH 2004, Lyon,
France, 6 10 June 2004.
Bixio, D., Parmentier, G., Rousseau, D., Verdonck, F., Meirlaen, J., Vanrolleghem, P.A. and Thoeye, C.
(2002). A quantitative risk analysis tool for design/simulation of wastewater treatment plants. Water Sci.
Technol., 46(4 5), 301 307.
Bode, H. and Lemmel, P. (2001). International product cost comparison in the field of water management.
Water Sci. Technol., 44(2 3), 85 93.
Claeys, F., Chtepen, M., Benedetti, L., Dhoedt, B. and Vanrolleghem, P.A. (2005). Distributed virtual
experiments in water quality management. Water Sci. Technol., 53(1), 297 305.
Copp, J., Spanjers, H. and Vanrolleghem, P.A. (eds) (2002). Respirometry in Control of the Activated Sludge
Process: Benchmarking Control Strategies IWA Publishing, London.
Gernaey, K., Rosen, C., Benedetti, L. and Jeppsson, U. (2005). Phenomenological modeling of wastewater
treatment plant influent disturbance scenarios. In: Proceedings of 10th International Conference on Urban
Drainage, Copenhaghen (Denmark), 21 26 August 2005.
Gillot, S., De Clercq, B., Defour, D., Simoens, F., Gernaey, K. and Vanrolleghem, P.A. (1999). Optimization
of wastewater treatment plant design and operation using simulation and cost analysis. In: Proceedings of
72nd Annual WEF Conference and Exposition, New Orleans, USA, 9 13 October 1999.
Henze, M., Gujer, W., Mino, T. and van Loosdrecht, M.C.M. (2000). Activated sludge models ASM1, ASM2,
ASM2d and ASM3, IWA Publishing, London.
Hulsbeek, J.J.W., Kruit, J., Roeleveld, P.J. and van Loosdrecht, M.C.M. (2002). A practical protocol for
dynamic modelling of activated sludge systems. Water Sci. Technol., 45(6), 127 136.
Langergraber, G., Rieger, L., Winkler, S., Alex, J., Wiese, J., Owerdieck, C., Ahnert, M., Simon, J. and
Maurer, M. (2004). A guideline for simulation studies of wastewater treatment plants. Water Sci.
Technol., 50(7), 131 138.
Melcer, H., Dold, P.L., Jones, R.M., Bye, C.M., Takacs, I., Stensel, H.D., Wilson, A.W., Sun, P. and Bury, S.
(2003). Methods for Wastewater Characterisation in Activated Sludge Modelling, Water Environment
Research Foundation, Alexandria, VA, USA - IWA Publishing, London.
Rousseau, D., Verdonck, F., Moerman, O., Carrette, R., Thoeye, C., Meirlaen, J. and Vanrolleghem, P.A.
(2001). Development of risk assessment based technique for design/retrofitting of WWTPs. Water Sci.
Technol., 43(7), 287 294.
Solvi, A.-M., Benedetti, L., Gille, S., Schlosseler, P., Weidenhaupt, A. and Vanrolleghem, P.A. (2005).
Integrated urban catchment modelling for a sewer-treatment-river system in Luxembourg. In: Proceedings
of 10th ICUD, Copenhagen, Denmark, 21 26 August 2005.
Vanrolleghem, P.A., Benedetti, L. and Meirlaen, J. (2005). Modelling and real-time control of the integrated
urban wastewater system. Environ. Model. Software, 20, 427 442.
Vanrolleghem, P.A., Insel, G., Petersen, B., Sin, G., De Pauw, D.J.W., Nopens, I., Weijers, S. and Gernaey, K
(2003). A comprehensive model calibration procedure for activated sludge models. In: Proceedings of
WEFTEC 2003: 76th Annual Technical Exhibition & Conference, Los Angeles, USA, 11 15 October
2003.
Vanrolleghem, P.A., Jeppsson, U., Carstensen, J., Carlsson, B. and Olsson, G. (1996). Integration of
wastewater treatment plant design and operation a systematic approach using cost functions. Water Sci.
Technol., 34(3 4), 159 171.