water research 44 (2010) 1765–1776

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Temperature as an alternative tracer for the determination

of the mixing characteristics in wastewater treatment plants

Markus Ahnert*, Volker Kuehn, Peter Krebs

Institute for Urban Water Management, Technische Universität Dresden, 01062 Dresden, Germany

article info abstract

Article history: The hydraulic characteristics, i.e. the residence time distributions, of wastewater treat-
Received 11 May 2009 ment plant reactors are usually determined using conventional tracers. This paper aims to
Received in revised form present an alternative approach based on wastewater temperature. The step in tempera-
16 October 2009 ture change, e.g. from stormwater events with cold rainwater, is used as a tracer signal.
Accepted 26 November 2009 The method was verified using lab- and pilot-scale tests that showed very good agreement
Available online 29 November 2009 of the time series estimated both with conventional and temperature tracer methods.
Results from lab-scale tests exhibit a zone with a minor water exchange of about 10% of the
Keywords: volume of all reactors, while the respective zone in the pilot-scale tests was estimated at
Activated sludge modelling about 30% of the total volume. The short-circuit flow was more than 50% of the inflow
Residence time distribution resulting from gaps between the walls inside the reactor cascade. An application example
Flow-through characteristics shows the importance of reliable residence time distribution underlying activated sludge
Tracer modelling and the uncertainty associated with neglecting the determination of appropriate
Temperature flow-through characteristics.
Dead zone ª 2009 Elsevier Ltd. All rights reserved.
Short-circuiting
Compartment model approach

1. Introduction 1.1. Historical background

The objective of wastewater treatment is to minimise the Historically, the hydrodynamics of wastewater treatment
impact of urban pollution on water bodies. An analysis of plants (WWTP) were of little interest due to the following facts:
hydrodynamics and mixing characteristics is essential for
the reliable, model-based prediction of a plant’s perfor-  Because of safety factors in the design process the resulting
mance, especially when the design of the activated sludge reactor volume is mostly larger than necessary.
reactors is different from standard cases. In case of  Design procedures are based on high input load values (e.g.,
a distinct deviation of the flow-through characteristics in Germany 85 percentile of the maximum daily inflow
from the ideal flow behaviour, short-circuiting and large hour; DWA, 2000) rather than on mean values, however,
zones with a reduced water exchange (‘‘dead zone’’) may these design loads are generally not met on most days.
result in increased effluent concentrations and reduced  Designs are based on key scenarios with low temperatures
efficiencies. that only occur on a few days a year.

* Corresponding author. Tel.: þ49 351 46337020; fax: þ49 351 46337204.
E-mail addresses: [email protected] (M. Ahnert), [email protected] (V. Kuehn), [email protected]
(P. Krebs).
0043-1354/$ – see front matter ª 2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.watres.2009.11.047
1766 water research 44 (2010) 1765–1776

 Distinct dynamics of diurnal inflow and loading conditions method. It is not guaranteed that the optimal substitute model
may be interpreted as the reason for critical process condi- is found for the reactor model representation because this
tions rather than unsatisfactory hydrodynamic conditions. model is pre-defined by the user. A priori knowledge of the
 Long hydraulic retention times lead to the assumption that hydraulic characteristics to be expected is necessary. This
there is no short-circuiting. concept is independent of flow conditions. It can be used for
 General operation problems may interfere with other constant or fluctuating flows.
influences, such as non-ideal flow conditions, and make it The calibration parameters of the substitute reactor model
difficult to identify the contribution of the various factors. provide valuable information about the hydrodynamic char-
acteristics of the reactor under observation. For example,
The last decades were characterised by an increasing use Vrabel et al. (1999) showed a good match of the modelling
of activated sludge models (ASMs) to describe the internal results for CMA for a complex closed vessel with multiple
processes of a WWTP. Within the process of model calibra- impellers. They compared their CMA results with CFD simu-
tion (see e.g. Langergraber et al., 2004), the underlying lations that greatly underpredict the rate of mixing resulting
hydraulics are seldom the object of calibration, i.e. a certain from flow patterns that were not considered in CFD codes.
number of completely stirred tank reactors (CSTRs) is Besides, there were CMA applications in the fields of ventila-
assumed without separate validation. Only when mixing tion, drinking water treatment, anaerobic digestion, waste-
problems in WWTP reactors become obvious, investigations water treatment with constructed wetlands and many
into the characterisation of underlying hydrodynamics are biological or chemical reactor systems that can be found in the
considered important. For that purpose, tracer tests are literature.
a widely used tool to characterise hydrodynamics, i.e. the An advanced CMA approach was presented by Hocine et al.
residence time distribution (RTD), which has been common (2008). They generated a so-called superstructure CMA model
practice in chemical reaction engineering for more than 70 within a framework of a nonlinear optimisation algorithm to
years (e.g., MacMullin and Weber, 1935). Generally, tracer describe the complex flow pattern.
tests are costly and labour-intensive; moreover their use in Currently, tracer tests are widely accepted to describe the
WWTP is critical in most cases because the substances used hydrodynamic behaviour in WWTP. Several successful
may be hazardous to water or because varying background examples are reported in the literature. Burrows et al. (1999)
concentrations of wastewater components have an inter- tested different approaches to build a model of a WWTP with 4
fering effect on the signal. parallel elongated lanes with surface aerators. They used
There are several ways reported in literature to get a model based on the Tanks-in-Series-Model, TISM (e.g. Lev-
a hydraulic model from a WWTP. Makinia and Wells (2000) enspiel, 1999) with some modifications to describe effects of
used tracer measurements to calibrate a hydraulic model with short-circuiting and dead zones. This approach showed
dispersive flow. With this approach they want to overcome a significantly better fit to the measured data than the others
the empirical arrangement of CSTR to reproduce the flow tested. De Clercq et al. (1999) used the TISM approach to
pattern. A similar approach is used by Potier et al. (2005) and analyse different types of WWTP (e.g. elongated aeration tank,
Lemoullec et al. (2008) to develop a hydraulic model of aerated circular oxidation ditch, and trickling filter) and found signif-
channel reactors. Another approach is used by Alex et al. icant amounts of ‘‘dead space’’ in the range of 10–20%. Kjell-
(1999, 2002). They adapted results from CFD simulations of strand et al. (2005) analysed tracer test results of
activated sludge tanks to generate a hydraulic model in a denitrification tank with the model used by Burrows et al.
a more structured way. Unfortunately there were no infor- (1999). As a result, a considerable short-circuit flow along the
mation about the calibration of the CFD simulations in the water surface was detected. These examples show the prac-
papers. ticability and the usefulness of tracer tests in WWTP.
Another technique that is widely used to interpret tracer
1.2. Alternatives in evaluation of tracer tests test results is the deconvolution of an input and an output
tracer signal in the frequency domain. It is based on the
Cholette and Cloutier (1959) presented some simple substitute concept of the convolution integral as described by Levenspiel
models for reactors to describe effects such as plug flow, (1999), but was established much earlier in signal and systems
short-circuiting and dead zones together with equations and theory. The advantage of this method is a result that is inde-
diagrams describing the characteristics of the reactor models. pendent of a given substitute model. It can be seen as a small
Milbury et al. (1965) refined the method to develop the so- disadvantage that the resulting function of the impulse
called Compartment Model Approach (CMA) which is based response of the reactor has to be analysed for irregular
on the idea that a reactor can be approximated by a simple- hydraulic characteristics (e.g. short-circuit and/or dead zones)
structure reactor model with internal flows and a more and ultimately needs to be transferred into a reactor model for
complex reactor system by a combination of linked simple modelling purposes anyway. A much greater disadvantage
reactor models. The mixing characteristics were diversified comes from the calculation of the deconvolution. This algo-
through model parameters that have to be calibrated to rithm is mathematically unstable, especially because of the
represent a real system. This technique has the advantage expected relatively low data quality of the tracer tests in the
that an appropriate model can be created on the basis of a real WWTP due to several influences, e.g. the problems of
reactor configuration and a parameter estimation procedure sampling spot and representativeness. Besides this, the flow
resulting in a set of parameters that are physically interpret- rate needs to be kept constant or flow fluctuation must be
able. At the same time, however, this is a disadvantage of the included into the calculation procedure.
 water research 44 (2010) 1765–1776 1767

As a descriptive example, Ramirez and Cortes (2004) pre- To establish a direct relation between a ‘real’ tracer test
sented results of some radiotracer investigations conducted with salt (sodium chloride) and temperature, a test solu-
for a certain reactor. From the results of the deconvolution tion with known salt concentration and respective
procedure they drew conclusions that the mixing behaviour is conductivity was cooled to 4  C and pumped as a Dirac-
random. Our own calculations using the same raw data and pulse to the first reactor. Before and after the tracer pulse,
a smoothing procedure adapted from Mills and Dudukovic the reactor cascade was fed with clear water at ambient
(1988) yield even better results. The resulting parameters temperature. Conductivity and temperature were logged
show a good correlation to the reactor characteristics with a high temporal resolution at different points of the
depending on different hydraulic loadings. Due to the various reactors.
disadvantages and uncertainties, the deconvolution method A series of four experiments was conducted. Due to
is not considered appropriate for analysing tracer test results different inflow rates, the mean hydraulic residence time
from WWTP. (HRT) was in the range of 1.6–1.9 h which is a smaller value
than that in the activated sludge tank of a real WWTP
1.3. Objective of this paper (typically 3–6 h) on the one hand side but still short accord-
ing to model-scaling laws. The maximum temperature
Independent of the method used for the development of the decrease during feeding with cold water in reactor 1 is in the
hydraulic model of a WWTP the basis is always a tracer test. range of 5.1–11.4  C which is more pronounced than in the
Because of the time and effort involved in using common real system. This leads to a stronger stimulus response from
tracer tests and the problems occurring in practical appli- cold water in the lab-scale reactor than from stormwater in
cations in WWTP, this paper describes the investigation of a real WWTP. The influence of the resulting stronger
an alternative approach. Most WWTP have a well developed density-effect in the lab-scale reactor is not as pronounced
data acquisition and storage system. From these routinely since the mixing is intensified through stirring. However, the
collected and most conveniently online measured data, the lab-scale experiment was not conducted to represent real
wastewater temperature was identified to be a suitable systems, but rather as a well defined test case for method
parameter to estimate the flow-through characteristics in development.
a WWTP designed for combined water treatment.
A combined water inflow induced by cold rainwater leads to 2.2. Pilot-scale system
a negative impulse response in the water temperature in
WWTP tanks and their effluents. The temperature signal can For scale-up validation, our pilot-scale WWTP was used. This
be used as an alternative to a tracer substance. This was plant is designed for approximately 40 PE, includes pre-
already presented in literature. Mayr et al. (1992) established denitrification and nitrification and was operated similarly to
the temperature pulse method to analyse the flow pattern in the Dresden WWTP and fed with the wastewater from the
chemical reactors. This method is used successfully by real plants influent. The inflow rate was set in relation to the
others too (e.g. Delvigne et al., 2005) but often not cited inflow of the large-scale plant resulting in a similar hydraulic
correctly. This paper describes the adaption resp. develop- retention time. The pilot-scale activated sludge system
ment of the temperature pulse method starting with lab- consisted of five cascaded reactors, separated by movable
scale experiments and a comparison with conventional walls, with a total volume of 6.6 m3 with a total reactor
tracer tests. The algorithm is then adopted and verified for height of 3.5 m. Thus the single cascades are high reactors
experiments in a pilot-scale WWTP. Finally, some practice- with a small surface and a large wall area to the next
relevant findings are discussed. Objective of this paper is the cascade. The scheme of the reactor is shown in Fig. 1 (right).
description of the method. For transparency reasons, simple The return activated sludge flow was controlled by the
hydraulic models are used. They could be replaced, however, process control system. Several online-sensors for various
by more complex models. parameters were also logging temperature so that tempera-
ture measurements are available from all 5 cascaded reac-
tors. Additional conductivity probes were installed for salt
2. Material and methods tracer detection.
For the pilot-scale system, a separate performance of salt
2.1. Lab-scale system and temperature tracer experiments was required. Conduc-
tivity tracer tests were performed in a special operation mode
The laboratory set-up consisted of three stirred and thus without any recycle flow in order to obtain more transparent
completely mixed reactors in series with a total volume of tracer information with regard to the flow conditions. The
about 10.3 l. No recycle flow was installed in order to prevent routine temperature data stored in the data acquisition
the signal intensity from being reduced by re-entrainment of system during real stormwater events (no additional cooling
tracer or cold water, respectively, i.e. the response signal of of the inlet) were used to analyse the temperature tracer tests.
a reactor cascade with recycle tends towards that of a fully From a time series of seven months duration, nine rain events
mixed single reactor and thus we would lose interpretability. were selected, inducing temperature differences between 1.6
A peristaltic pump was used for feeding. All reactors and the and 4.6  C in the first cascade. The mean HRT was in the range
feeding system were checked with regard to volume and flow of 5.6–10.1 h. The specific feeding rate was in the same order of
rate. All tracer tests were performed regarding to Levenspiel magnitude as for a large-scale plant in normal operation
(1999) and AWWA (1996). mode.
1768 water research 44 (2010) 1765–1776

Q·fSC
Influent Effluent
Influent Q·(1-fSC) Effluent

V·factive

V·(1-factive)

Fig. 1 – Scheme of substitute hydraulic model (left) with short-circuit flow Q$fSC and inactive reactor volume V$(1 L factive)
and scheme of cascade reactor for pilot-scale tests (right).

2.3. Procedure 2.3.2. Energy balance

The calculation of the overall energy balance of the water in
2.3.1. Substitute hydraulic model the reactor is necessary for using the wastewater temperature
Each spatially differentiated reactor (ideally a completely as an alternative tracer. In general, the following fluxes may
stirred tank reactor CSTR) was replaced by a substitute significantly influence the energy in the effluent of the reactor
hydraulic model reactor containing the main non-ideal (adapted from La Cour Jansen et al., 1992):
hydraulic characteristics which are a short-circuit flow
directly into the effluent and a dead zone inside the reactor  Influent energy flux (represented by water temperature)
(Fig. 1 left). This simple substitute model was chosen to get Q_ Infl
a simple optimisation algorithm for parameter estimation  Energy flux from biological processes (exothermic reactions)
because this paper presents only the method to use the Q_ Bio
temperature instead of a real tracer. Other possible substi-  Energy flux from radiation (short- and long-wave radiation)
tute models can be found in hydraulic literature (e.g. Lev- Q_ Rad
enspiel, 1999) or in the field of practical analysis of flow  Energy flux from sensible heat (wind-dependent exchange
pattern in WWTP to develop a hydraulic model (e.g. Burrows over surface) Q_ Sens
et al., 1999; Alex et al., 2002; Kjellstrand et al., 2005). These  Energy losses due to evaporation Q_ Evap
models may be more suitable to describe the hydraulics in  Energy from friction losses of pumps and from aeration unit
WWTP reactors but for method development they seem to Q_ frict þ Q_ aer .
be inappropriate.
Two parameters are to be calibrated. The short-circuit Some simplifications could be made because for the indoor
factor fSC is the flow fraction reaching the effluent more or less test site conditions the following of the above mentioned
directly from the influent. The active volume factor factive is energy contributions are negligible: (i) heat transfer due to
the fraction of the total reactor volume where water exchange radiation because of small temperature differences between
and mixing are significant while fraction (1  factive) corre- the reactor and indoor atmospheric conditions, (ii) heat
sponds to the inactive volume in the reactor. The thorough transfer due to evaporation and (iii) sensible heat transfer
assumption that a dead zone V$(1  factive) is completely de- because of small temperature differences, the absence of air
coupled from the flow through the reactor is – for an open movements and high relative humidity in the pilot-plant hall.
vessel – impossible both in theory and in real systems Accordingly, all energy exchanges with the environment for
(however, bulk fluid and interstitial water in colmated sedi- indoor reactors were condensed in the calculation of energy
ments may come close to this assumption). Instead, the convection Q_ conv .
inactive zone should slightly interact with the active zone, With these modified contributions, the overall energy
which in the model could be described by an exchange rate balance can be established (Equation (1)).
between the two zones, where inflow and outflow are only
connected to the active zone. This leads to a model with an Q_ Effl ¼ Q_ Infl þ Q_ Bio þ Q_ conv þ Q_ frict þ Q_ aer (1)
additional parameter to be calibrated. Thus, for reasons of The influent heat energy flux Q_ Infl was calculated using the
transparency, the simpler conceptual approach with a dead measured inflow temperature. The heat calculated from the
zone was chosen for this work. The two parameters fSC and effluent temperature was used to optimise the other heat
factive are assumed independent of variations of the flow rate. energy terms discussed above.
Due to the perfect separation of the reactors in the lab- The convective energy exchange Q_ conv was calculated
scale cascade, i.e. short-circuit flow was physically impos- according to Equation (2). The heat transfer depends on the
sible, the respective substitute model was simplified. The total reactor plus water surface area A, the temperature
remaining calibration parameter was the fraction of the difference DT and a constant k representing material proper-
active reactor volume factive. In the pilot-scale experiments, ties (Long, 1999). For simplification, surface area and constant
physical short-circuit flow was obvious due to gaps between k were condensed in parameter fHET.
the side walls and the lateral baffles. Accordingly, the short-
circuit factor fSC was additionally implemented in the model. Q_ conv ¼ k$A$DT ¼ fHET $DT (2)
 water research 44 (2010) 1765–1776 1769

The reliable quantification of the contributions of aeration, In all following comparisons between measured and esti-
pumps and biological processes is quite difficult. Normally, mated tracer responses, three different goodness-of-fit
energy efficiency data of the technical units are not available criteria were used:
and the calculation of the energy flux from the biological
conversion processes would require detailed energy-conver-  Root mean square error (RMSE) as a unit-dependent
sion data of the substances involved. Therefore, a simplified measure can be used to establish a relation between the
assumption was made to consider these temperature-influ- mean deviation and the mean measured values.
encing factors: The energy fluxes resulting from aeration,  Coefficient of determination (R2) is a widely used correla-
technical units and biological processes depend on the tion-based criterion.
influent flow rate as an indicator for load variations, which is  The coefficient of efficiency E is a difference-based good-
a reference for the change of energy production as a result of ness-of-fit approach that is widely accepted in the field of
biological processes, aeration and also the energy flux hydrological modelling and has some advantages over other
resulting from pumping, since the return activated sludge criteria. It was used in a modified form with absolute rather
flow rate is controlled inflow-dependently. than squared differences due to the bigger sensitivity (E (E2 in
Q_ bioþaerþfrict was related to the flow rate Q in the reactor by this paper due to differentiation to modified form) originally
a constant coefficient fDET (Equation (3)). published by Nash and Sutcliffe (1970); E1 modifications
described in Ahnert et al. (2007)). E reached 1 for a perfect fit
Q_ bioþaerþfrict ¼ fDET $f ðQÞ (3)
between measurement and model result, 0 when the abso-
The energy equation for the lab-scale experiments could be lute sum of the differences between measured and esti-
further simplified since aeration, biological processes and mated values was as large as the variability in the measured
recycle flows were absent. The constant coefficients fHET and data. In case of negative E values, the measured mean was
fDET were used in the optimisation routines to adjust the a better predictor than the model.
measured to the estimated temperature data as described
below. The temperatures in the individual parts of the Equation (4) can be used for the calculation of both the
substitute model can be different. This is set as a boundary original ( j ¼ 2) and the modified ( j ¼ 1) form
condition depending on the number of available data points PN
from measurement. jMi  Ei jj
Ej ¼ 1  PNi¼1 j
(4)
i¼1 jMi  Mj

2.3.3. Parameter estimation with M, measured time series; E, estimated time series; and N,
All calculations were performed with Matlab 7.0.4 (The
number of data points.
Mathworks Inc.). The goal of optimisation was the mini-
misation of a selected goodness-of-fit criterion. First, the root
mean square error (RMSE) between the measured and the
estimated effluent temperature and conductivity, respec- 3. Results and discussion
tively, was selected. However, with respect to the results
discussed in the sensitivity test in the chapter below, a modi- In the following section, the results of conventional (conduc-
fied version of the coefficient of efficiency E (Nash and Sut- tivity) and alternative tracer tests (temperature) are discussed
cliffe, 1970) was used. The optimisation was performed with based on both lab-scale and pilot-scale experiments.
the Matlab-function fminsearch. It incorporates the simplex
search algorithm of Lagarias et al. (1998). 3.1. Lab-scale system
The estimation of the lab-scale experiments was adjusted
to each reactor separately. Two parameters had to be cali- A series of four runs with cold saltwater pulses was conducted
brated for the temperature-based analysis: the active volume as described above. Each reactor of the lab-scale cascade was
fraction factive and the parameter for the heat energy transfer optimised separately with the selected substitute hydraulic
fHET. For the conductivity-based calculations, only the active model (Fig. 1). The resulting parameters (mean values from the
volume fraction factive had to be calibrated. four runs) for the mean active fraction factive are summarised in
Parameters for the pilot-scale conductivity tracer experi- Table 1. An excellent agreement between measured and esti-
ments were estimated for all cascaded reactors simulta- mated values was found, which is illustrated by the different
neously with the same set of parameters for short-circuiting goodness-of-fit measures. R2 is greater than 0.99 in all cascade
flow fraction fSC and active volume fraction factive. Because all reactors. The predictions also proved to be correct with the
reactor units had the same dimensions, we assumed that the modified coefficient of efficiency E1 with values greater than
hydraulic characteristics were also similar. No distinction was 0.9. In Fig. 2, measurements and simulations of an experi-
made between aerated and nonaerated cascades under the mental run are shown in all three reactors for both types of
assumption that no significant differences in mixing occur in tracer tests where measured and estimated lines are almost
these two types of cascades. Two more parameters had to be covering. The input pulse of cold saltwater is clearly repre-
optimised for the temperature tracer estimations: the heat sented by the measurements in the inlet. The increasing air
transfer fHET to and from the atmosphere through the inter- temperature at the test site from 23  C to 25.5  C during the
face area (water surface and reactor walls) and the dynamic experiment was considered as a dynamic boundary condition.
heat transfer fDET through pumps, aeration and biological The resulting effects of convection were modelled well, which
reactions. is reflected by the increasing effluent temperature tendency of
1770 water research 44 (2010) 1765–1776

Table 1 – Optimisation results for the modelled reactors of the lab-scale system.
cascade no. mean active volume fraction factive mean relative deviation goodness-of-fit measures

mean RMSE mean R2 mean E1

conductivity tracer
R1 0.936 0.051 32.8 0.995 0.950
R2 0.894 0.054 25.4 0.999 0.953
R3 0.857 0.043 20.2 0.999 0.941

temperature tracer
R1 0.897 0.069 0.13 0.997 0.937
R2 0.844 0.095 0.08 0.998 0.944
R3 0.854 0.097 0.06 0.996 0.944

reactor R3. For reactors R1 and R2, the temperature-based tracer tests on the biocenosis, these tests were performed
estimations of the active volume are some 5% lower than the before moving to regular operating conditions. Thus the
conductivity-based estimations (Table 1). For cascade R3, the conductivity and temperature tracer tests were not conducted
estimation is almost identical for both methods. From the low simultaneously. A series of three conductivity tracer tests was
relative deviation of the ratio between the mean value and the performed. Table 2 shows the optimised parameters of the
standard deviation of 4.3–5.4% for conductivity tracer and 6.9– structured model (Fig. 1). The mean value of the active volume
9.7% for temperature tracer it can be concluded that the two fraction factive is about 0.717. This relatively small value, i.e.
methods suit each other well, while the relative deviation is a relatively large dead zone, can be attributed to the geometry
slightly higher for temperature tracer results. The short-cir- with a small surface area in relation to the depth. The amount
cuiting ratio fSC was set to nearly zero by the optimisation of short-circuiting is surprisingly high. Investigations made
algorithm due to the complete partitioning of the three reac- after the tracer tests lead to the conclusion that short-cir-
tors, when fSC was integrated in the optimisation routine. cuiting is caused by the walls between the reactors. Gaps
The active volumes identified correspond to a total dead between the outer wall and the internal walls separating the
zone of about 10–15% in all reactors. Due to the relative reactors allow for a direct flow-through along the outer wall.
deviation between the different experiments, a smaller Thus, the estimated short-circuit factor of 0.568 appears to be
amount of dead zone may be unidentified independently of realistic. As an example, Fig. 3 (right) shows the conductivity
the tracer type used. in the effluent of the 5th reactor, where the measured and
A statistical test (Student’s t-test with a ¼ 0.05) showed that estimated time to the conductivity peak was nearly the same.
the estimations made with the two methods for all reactors When five CSTRs with full volume were assumed for the
show no significant difference in the mean values of the single cascade, the peak occurred later and was more pronounced
experiments. Furthermore, the comparison of the active due to the absence of short-circuit flows or a smaller active
volume fraction approach and the hypothesis of an ideal CSTR volume, respectively.
showed statistically significant differences for all three reac- In the pilot-scale test simulations, the goodness-of-fit
tors and both tracer methods (a ¼ 0.05). Thus the estimation of measures and especially the coefficient E1 exhibited a lower
a dead zone of about 10–15% is statistically significant. quality of fit than in the lab-scale test simulations. However,
the quality is still good with R2-values greater than 0.94, while
3.2. Pilot-scale system the more significantly reduced E1-values prove to be more
sensitive. One reason for the lower model fit quality is the
A similar procedure was applied to the experiments with the application of an identical parameter set for all reactors. The
pilot-scale WWTP. Due to possible adverse effects of the salt use of individual model parameters for the five reactors in

26 7000
24 R1 meas. R1 est.
6000
Conductivity [μScm-1]

22 R2 meas. R2 est.
Temperature [°C]

20 5000 R3 meas. R3 est.

R1 Inflow
18 R1 meas. R1 est. 4000
16 R2 meas. R2 est. 3000
14 R3 meas. R3 est.
2000
12 R1 Inflow Air Temp.
10 1000
8 0
Time4:00
0:00 0:30 1:00 1:30 2:00 2:30 3:00 3:30 [h] 0:00 1:00 2:00 3:00 4:00 Time
5:00[h] 6:00

Fig. 2 – Example of a test event of temperature (left) and conductivity (right) measurements and estimations with selected
substitute hydraulic model in the effluent of the lab-scale system.
 water research 44 (2010) 1765–1776 1771

Table 2 – Optimisation results for the modelled reactors of the pilot-scale system.
temperature difference (K) event no active volume fraction bypass flow goodness-of-fit measures
fraction of inflow
RMSE R2 E1

conductivity tracer
#1 0.684 0.734 51.4 0.950 0.691
#2 0.698 0.510 60.7 0.948 0.725
#3 0.770 0.460 11.9 0.988 0.880

mean 0.717 0.568

std. dev. 0.038 0.119
rel. dev. 0.053 0.210

temperature tracer
4.3 #1 0.655 0.348 0.04 0.999 0.975
4.4 #2 0.873 0.416 0.09 0.996 0.935
4.5 #3 0.600 0.547 0.06 1.000 0.964
4.6 #4 0.747 0.556 0.09 1.000 0.956
1.9 #5 0.634 0.645 0.08 0.993 0.915
1.6 #6 0.628 0.571 0.06 0.996 0.936
2.5 #7 0.767 0.548 0.13 0.986 0.879
3.0 #8 0.675 0.485 0.15 0.976 0.805
3.3 #9 0.712 0.589 0.05 0.998 0.963

mean 0.699 0.523

std. dev. 0.081 0.087
rel. dev. 0.116 0.166

series would lead to 12 instead of 4 parameters to be cali- The routine data from the plant operation were used not
brated. This would result in an even more pronounced over- only for the conductivity tracer tests but also as the source of
parametrisation with the consequence of much longer temperature tracer data. Nine rain events with inflow
calculation times. temperature differences between 1.6 and 4.6 K were extrac-
A more compelling reason for the deviations between the ted from the database. The durations of these events were in
measured and the estimated effluent conductivity is the the range of 1.5–4 days. The resulting model parameters
model structure used. As some test simulations show, a model factive and fSC are shown in Table 2. An example run is
with a zone of reduced water exchange yields better results depicted in Fig. 3 (left). The mean active volume fraction of
than a model with a dead zone. The deviation can successfully the nine events is 0.699 and thus close to the value obtained
be reduced in particular in the tail of the conductivity (see from the conductivity tracer test estimation. Also, the
deviations in Fig. 3, right). Such a substitute model would resulting short-circuit factor of 0.568 (mean value) is in good
further increase the number of parameters. Moreover, agreement with that obtained from conductivity tests. The
parameter estimations with this more complex model relative deviations of the active volume fraction are low for
produced almost the same results for active volume fraction both methods with 5.3% for conductivity and 11.6% for
factive and short-circuiting rate fSC as with the dead-zone temperature tracer tests. The relative deviation for the short-
model. Thus, the use of the simpler dead-zone model with circuit factor varies in a wider range with 21% for the
only one parameter set for all reactors is assumed to be conductivity method and 16.6% for the temperature method.
sufficiently accurate also in view of the fewer measurement Possible reasons can be temporary blockage of the gaps
points for temperature tracer in normal plant operation. between the outer and internal walls due to clogging with

18 measured 1000
estimated 900 measured
Conductivity [μScm-1]

17 estimated (perfect Cascade) estimated

800
Temperature [°C]

700 estimated (perfect Cascade)

16
600
15 500
400
14 300
13 200
100
12 0
0 1 2 3 Time
4 [d] 5 0.0 0.5 1.0 1.5 Time
2.0[d] 2.5

Fig. 3 – Example of a rain event of temperature (left) and conductivity (right) measurements and estimations with ideal and
selected substitute hydraulic model in the effluent of the pilot-scale system.
1772 water research 44 (2010) 1765–1776

Table 3 – Kinetic parameters affecting nitrification.

kinetic parameters: source autotrophic maximum oxygen substrate saturation aerobic endogenous
growth rate mA (d1) constant KA,O (g O2 m3) respiration bA,O2 (d1)

original ASM3 (Gujer et al. 1999) 1.0 0.5 0.15

Koch et al. (2000) 1.3  0.4 0.5 0.2
calibration with realistic hydraulics 1.0 0.5 0.2
calibration with ideal hydraulics 0.72 0.55 0.24

ratio realistic/ideal hydraulics 0.72 1.1 1.2

sludge. The values of the goodness-of-fit criteria reveal minima. However, with the altered parameters (0.9factive and
excellent values for all events, though. As can be seen in 1.1factive), a similarly good agreement between modelled and
Fig. 3 (left), the estimated effluent temperature fits the estimated values was obtained resulting in small differences
measured value extremely well, whereas the assumption of in columns in Table 4 for each cascade. Moreover, according
CSTRs for all reactors instead of the structured model to the criterion R2 in three cases the optimised factive turns
exhibited a clear deviation between simulation results and out to be only second best while 1.1factive yields slightly
measurements. better fits (see numbers in underline). Apparently, this
A statistical analysis (Student’s t-test with a ¼ 0.05) of the criterion does not prove to be sufficiently sensitive for this
significance of the results showed similar findings as for the type of application.
lab-scale experiments and confirmed that there are no Due to the minor relative differences between the param-
significant differences between the mean values of the opti- eters from the optimisation algorithm and those obtained by
mised parameters in both methods (conductivity and variation by 10% of the goodness-of-fit criteria, a Student’s
temperature method) with respect to the varying number of t-test (a ¼ 0.05) was made and the results obtained were
experiments conducted for each method. compared for each criterion. For test calculations, all good-
ness-of-fit measures for the optimal parameter and the varied
3.3. Sensitivity tests for optimised substitute model parameter values were used. Significant differences between
parameters the goodness-of-fit criteria were obtained for the best fitting
parameter factive and both variants (0.9factive and 1.1factive) for
3.3.1. Lab scale tests the temperature tracer results. In case of conductivity tracer,
The applicability of the investigated tracer methods was there were significant differences only for the criteria R2 and
successfully demonstrated. Nevertheless, it is important to E1, but not for RMSE. A possible reason for this are the more
evaluate the factors that influence the results, such as pronounced dynamics in the time series of water temperature
measurement uncertainties or inadequate substitute models. compared to the salt tracer signal. These findings lead to the
The goodness-of-fit measures of the optimised parameter sets conclusion that the coefficient of efficiency is the best suited
all show an excellent agreement between measured and criterion for this kind of application and the use within opti-
estimated time series (Table 1). misation algorithms.
Some test simulations were performed by quantifying the
goodness-of-fit criteria for a model parameter factive that was 3.3.2. Pilot scale tests
varied by 10% from the optimal value. The resulting A similar procedure was applied to the simulations of the
goodness-of-fit measures are listed in Table 4. For all reac- pilot-scale temperature experiments. In summary, no signif-
tors and both kinds of tracers, the best fit (italic) according to icant differences were found between the original and the
RMSE and E1- and E2-criteria was obtained from the simu- varied values of the parameter factive according to the evalu-
lations with the original values of factive, which should be the ated goodness-of-fit criteria. Still, the goodness-of-fit criteria
case with automated parameter estimation without local were best for the original factive value in all cases. Possible

6 estimated (real Cascade) estimated (ideal Cascade) 35

online measurement 24h composite sample
ammonia nitrogen [gm-3]

30
nitrate nitrogen [gm-3]

5
25
4
20
3
15
2
10 estimated (real Cascade) estimated (ideal Cascade)
1 5 24h composite sample

0 0
0 2 4 6 8 10 12 [d] 14
Time 0 2 4 6 8 10 Time
12 [d] 14

Fig. 4 – Ammonia (left) and nitrate (right) nitrogen test simulation for pilot-scale system for realistic and ideal cascade with
identical parameter set.
 water research 44 (2010) 1765–1776 1773

is nearly a complete match of all three criteria as can be seen

Table 4 – Goodness-of-fit measures from sensitivity
analysis of lab-scale tests. from the marked points in all 3 plots.
Another important finding from the contour plots in Fig. 6
cascade test case RMSE R2 E2 E1
is that the sensitivity of the result on the two model param-
conductivity tracer eters is distinctly different. The elliptic area of best value is
R1 factive from optimisation 32.7 0.9946 0.9931 0.950 between a small band of the active volume fraction factive but
0.9 factive 101.6 0.9896 0.9889 0.931
covers a much wider band of the short-circuit factor fSC. We
1.1 factive 84.2 0.9961 0.9929 0.935
conclude that this is the reason for the large variability in the
R2 factive from optimisation 25.3 0.9991 0.9969 0.953 fSC-values for the different experiments (see Table 2). The
0.9 factive 50.7 0.9962 0.9903 0.919 horizontally situated elliptic area of best fit in Fig. 6 shows no
1.1 factive 47.5 0.9964 0.9914 0.918
correlation between both parameters. This leads to the
R3 factive from optimisation 20.2 0.9986 0.9959 0.941 conclusion that there is no dependency of both parameters.
0.9 factive 34.5 0.9952 0.9899 0.910
1.1 factive 29.5 0.9994 0.9929 0.936
3.4. Influence of recirculation
temperature tracer
R1 factive from optimisation 0.130 0.9975 0.9957 0.937
As described above the tracer tests for lab scale were per-
0.9 factive 0.174 0.9949 0.9930 0.921
1.1 factive 0.153 0.9979 0.9948 0.927
formed without any recirculation just like the salt tracer test
at pilot scale. This was done to avoid perturbation with tracer
R2 factive from optimisation 0.077 0.9976 0.9970 0.944
material as it can be found in literature (e.g. Tizghadam et al.,
0.9 factive 0.116 0.9958 0.9937 0.925
2008). Only the temperature tracer data at pilot scale were
1.1 factive 0.119 0.9960 0.9932 0.917
measured with recirculation due to the fact that this was done
R3 factive from optimisation 0.057 0.9964 0.9951 0.944
under normal working conditions. It is possible that a signifi-
0.9 factive 0.078 0.9928 0.9920 0.927
cant recirculation flow can influence the hydraulic charac-
1.1 factive 0.180 0.9958 0.9850 0.893
teristics. In case of pilot-scale tests no differences were found
between the two types of tracer data that can be tracked back
to the differences in recirculation. Since the integration of
reasons are the high dynamics of the temperature measure-
recirculation in the described method resp. the data used for
ments and the higher uncertainty in the measured values due
parameter estimation is basically possible it should be
to the larger scale of the plant. The conductivity tracer tests
considered to avoid misinterpretation of the results.
were not evaluated statistically due to the small number of
experiments.
The sensitivity of both parameters factive and fSC was tested 3.5. Consideration of variable flow rates
by running a series of simulations and varying both model
parameters in a range between 70% and 130% around the Tracer tests usually were performed with constant flow
value obtained by automated parameter optimisation. It was rate. Main reason is the simplicity of interpretation and use
performed using the measurements of the experiment of equations based on RTD theories with constant flow rate.
demonstrated in Fig. 3. The three goodness-of-fit criteria are In WWTP under normal working conditions the flow rate
depicted as contour plots in Fig. 6. The area of best fit is fluctuates in daily cycles and also in stormwater conditions.
smallest for coefficient of efficiency E1, i.e. it is most sensitive. The use of temperature as a tracer requires important drops
Therefore, this criterion was used as the optimisation goal in due to cold stormwater events. In this case the flow rate is
the parameter estimation algorithm as discussed above. normally not constant and the hydraulic characteristics
Moreover, the slope in the outer regions of the contour plots may vary compared to dry weather conditions. The
for R2 and E1 is steeper than for RMSE, which indicates accel- simplest way to prevent this problem would be to limit the
erated gradient-based parameter optimisation algorithms. flow rate in the range of dry weather flow. This could be
The point of best fit varies slightly for fSC, while for factive there done by using storage basins to buffer the additional inflow.

7 estimated (real Cascade) 35

est. optimised (ideal Cascade)
online measurement
ammonia nitrogen [gm-3]

6 30
nitrate nitrogen [gm-3]

24h composite sample

5 25
4 20
3 15
2 10 estimated (real Cascade)
estimated optimised (ideal Cascade)
1 5 24h composite sample

0 0
0 2 4 6 8 10 Time
12 [d] 14 0 2 4 6 8 10 Time
12 [d] 14

Fig. 5 – Ammonia (left) and nitrate (right) nitrogen test simulation for pilot-scale system for realistic and ideal cascade with
optimised parameter set (for ideal cascade).
1774 water research 44 (2010) 1765–1776

RMSE E1 R2
0.85

0.80

0.75
factive [−]

0.70
(0.355;0.655) (0.359;0.657)
0.65
(0.348;0.655)
0.60

0.55

0.50
0.25 0.30 0.35 0.40 0.45 0.25 0.30 0.35 0.40 0.45 0.25 0.30 0.35 0.40 0.45
f [−] f [−] f [−]
SC SC SC

Fig. 6 – Contour plots for the different goodness-of-fit measures for varied parameters of the substitute hydraulic model.

The resulting hydraulics according to dry weather denitrification plant with two mixed and three aerated
conditions. compartments at a sludge age of 15 d and a wastewater
Beside this there exist developments in RTD theory that temperature of about 20  C.
consider variable flow rates. First work was done by Nauman Using the parameter set of Koch et al. (2000), the plant was
(1969). Niemi (1977) continues by replacing the fixed reference modelled with the structured hydraulic model and the mean
time with flow-weighted time. His mathematical fundamen- parameters for factive and fSC on the one hand, and with 5 ideal
tals in this domain were basis for practical application (e.g. CSTRs in series on the other hand. The results for ammonia
Furman et al., 2005; Werner and Kadlec (2000)). They showed and nitrate nitrogen effluent are shown in Fig. 4. The NH4-N
the practical reproducibility of tracer tests with constant and effluent concentration estimated with the structured
variable flow rate resulting in identical results from both approach is in good agreement with the online measurement.
methods. Thyn and Zitny (1996) restricted the use of the In contrast, the estimation with ideal CSTRs shows a poor fit
method with variable flow rate to cases were the hydraulic and the values are at a considerably lower level, which can be
characteristics are not changed by fluctuations in flow rate explained by the absence of a short circuit and the associated
(e.g. number of CSTR). Leclerc et al. (2000) confirm that RTD absence of poorly treated flow fractions. As a consequence,
theory is valid at least for low-complex systems with variable nearly all ammonia is oxidised, which causes nitrate effluent
flow rate. The validity for very complex systems needs to be concentrations that are higher than those estimated with
verified. Calculations of the authors with the algorithm pre- structured hydraulics.
sented by Furman et al. (2005) showed feasible results in Without information on the hydraulic characteristics, the
WWTP application. common approach to improve the agreement between simu-
lation and measurement would be to modify some kinetic
3.6. Impact of the structured model approach parameters, still based on the ideal CSRT model. Here, this
on ASM parameters was done for the kinetic parameters that affect the nitrifying
biomass (Table 3). Fig. 5 shows the improved agreement
The findings described above were used to analyse the impact obtained with the calibrated kinetic parameters which now
of the adapted reactor approach as compared to the are of nearly the same quality as for the structured hydraulic
assumption of ideal CSTRs. In most cases of ASM modelling approach. Nevertheless, there is a small time shift in peak
for consultancy, the budget does not allow for experiments for concentrations resulting from the absence of short-circuiting
parameter estimation and identification of hydraulic charac- flow and the higher active volume. The maximum autotrophic
teristics. Instead, a set of commonly approved standard growth rate mA was reduced to 72% of the value applied with
parameters is applied (e.g., Bornemann et al., 1998, for ASM1 structured hydraulics approach, where the general recom-
(Henze et al., 1987); Koch et al., 2000, for ASM3 (Gujer et al., mendation of Gujer (1999) was used. This value is low but in
1999)). The reactor configuration is derived from geometrical a possible range for a system with inappropriate nitrification
boundary conditions and the segmentation of large aeration capacity, e.g. due to some inhibitory effects. Further, the
tanks into a CSTR-cascade is mostly done intuitively. oxygen saturation constant KA,O and the aerobic endogenous
As a test example, the previously described pilot-scale respiration rate bA,O are higher by 10% and 20%, respectively,
system was used for modelling the processes with ASM3 than the recommended values applied on the structured
(Gujer et al. 1999). A time series of 16 days was selected. Two- hydraulics basis. Altogether, the nitrifying biomass in case of
hour composite samples of COD and total Kjeldahl Nitrogen CSTR hydraulics has inappropriate growth conditions and the
(TKN) were collected from the influent. In the effluent, online- nitrification capacity is strongly underestimated. The cali-
measurements (ammonia nitrogen) were performed and 24- brated model may well fit the measurements in the calibration
h-composite samples (ammonia and nitrate nitrogen) were period but can lead to erroneous predictions and conclusions
analysed. The pilot WWTP was operated as a pre- for different operation conditions and design purposes. In the
 water research 44 (2010) 1765–1776 1775

latter case, the necessary aerobic volume of an activated Finally, temperature is not the only conceivable alternative
sludge reactor without short-circuit would be over predicted tracer measurement. In some special cases, ammonia or
by more than 30%. nitrogen can also be used (e.g. Braun and Gujer, 2008). Under
the assumption of no conversion processes in primary or
secondary settlers, the measurements of sufficiently fast
4. Conclusions online probes in influent and effluent can be used for a similar
evaluation. If the conversion processes are integrated into this
This paper presents a method to use inflow and effluent methodology, reactive tracers may also be used.
temperature as a tracer signal to estimate the hydraulic
characteristics of a reactor. This method is applicable inde- references
pendent of the used hydraulic model. There should be an
agreement between the complexity of the hydraulic model
resp. the great numbers of parameters to estimate and the Ahnert, M., Blumensaat, F., Langergraber, G., Alex, J., Woerner, D.,
available measured data. Frehmann, T., Halft, N., Hobus, I., Plattes, M., Spering, V.,
In combined sewer systems under central European Winkler, S., 2007. Goodness-of-fit measures for numerical
modelling in urban water management – a review to support
conditions, the inflow temperature changes during rain
practical applications. In: Proceedings LWWTP07 Conference,
events can be interpreted as alternative tracer experiments.
Vienna, Austria, pp. 69–72.
The tests of the method introduced showed promising results Alex, J., Tschepetzki, R., Jumar, U., Obenaus, F., Rosenwinkel, K.H.
for both lab- and pilot-scale experiments and thus the method , 1999. Analysis and design of suitable model structures for
has the potential to reduce costs for identifying hydraulic activated sludge tanks with circulating flow. Water Science
characteristics. Besides, it prevents misuse of calibration in and Technology 39 (4), 55–60.
over-parametrised models, e.g. activated sludge models. The Alex, J., Kolisch, G., Krause, K., 2002. Model structure
identification for wastewater treatment simulation based on
method can be applied to all kinds of flow conditions with real
computational fluid dynamics. Water Science and Technology
plant data. The plant or reactor under investigation does not 45 (4–5), 325–334.
need to be running in a constant flow mode under assumption AWWA, 1996. Tracer Studies in Water Treatment Facilities: A
of only minor influence on hydraulic characteristics. Protocol and Case Studies. American Water Works
It must be noted that the estimated amount of short-cir- Association Research Foundation.
cuiting may be less accurate compared with the estimated Bornemann, C., Londong, J., Freund, M., Nowak, O., Otterpohl, R.,
active volume of the investigated reactors. For generalisation, Rolfs, T., 1998. Hinweise zur dynamischen Simulation von
Belebungsanlagen mit dem Belebtschlammodell Nr. 1 der
more systems have to be investigated. At least the active
IAWQ (Remarks to dynamic simulations of activated sludge
volume is the more sensitive parameter to calibrate. These systems with Activated Sludge Model No. 1 of the IAWQ).
findings are model-specific. It is assumed that the use of Korrespondenz Abwasser KA 45 (3), 455–462 (in German).
another substitute model would give complete other inter- Braun, D., Gujer, W., 2008. Reactive tracers reveal hydraulic and
pretation of results. control instabilities in full-scale activated sludge plant. Water
The results obtained in investigations in lab-scale tests Science and Technology 57 (7), 1001–1007.
Burrows, L.J., Stokes, A.J., West, J.R., Forster, C.F., Martin, A.D.,
lead to dead zones of about 10–15% of the reactor volume.
1999. Evaluation of different analytical methods for tracer
Although these amounts are small, they are statistically
studies in aeration lanes of activated sludge plants. Water
significant. The developed method is suitable for such small Research 33 (2), 367–374.
deviations from ideal hydraulic conditions. The same can be Cholette, A., Cloutier, L., 1959. Mixing efficiency determinations
concluded for pilot-scale tests were the active volume fraction for continuous flow systems. Canadian Journal of Chemical
is much smaller with values around 0.7 and a short-circuiting Engineering 37, 105–112.
flow of about 50% of the inflow. De Clercq, B., Coen, F., Vanderhaegen, B., Vanrolleghem, P.A.,
1999. Calibrating simple models for mixing and flow
For the application in large WWTP it is necessary that
propagation in wastewater treatment plants. Water Science
a separate analysis of single reactors is carried out instead of and Technology 39 (4), 61–69.
considering the complete reactor system with all internal Delvigne, F., Destain, J., Thonart, P., 2005. Structured mixing
recirculation. The resolutions of the temperature measurement model for stirred bioreactors: an extension to the stochastic
in time and scale as well as the number of measurements in the approach. Chemical Engineering Journal 113 (1), 1–12.
reactors are important for the quality of the results. In most of DWA, 2000. ATV-DVWK Arbeitsblatt A 131: Bemessungen von
the commonly used measuring devices, the temperature is einstufigen Belebungsanlagen ab 5000 EW, Design guideline
for single-stage activated sludge systems with more than 5000
standardly metered along with the actual measurement and can
PE. DWA, Hennef, Germany (in German).
be easily logged in the data acquisition system of the plant. Furman, L., Leclerc, J.P., Stegowski, Z., 2005. Tracer investigation
A descriptive example was used to show that neglecting the of a packed column under variable flow. Chemical Engineering
hydraulic characteristics in an activated sludge model can lead Science 60 (11), 3043.
to modelling results that were obviously acceptable because Gujer, W., Henze, M., Mino, T., van Loosdrecht, M., 1999. Activated
the calibrated model parameters are in a theoretical possible sludge model No. 3. Water Science and Technology 39 (1), 183–193.
Henze, M., Grady, C.P.L., Gujer, W., Marais, G.V.R., Matsuo, T.,
range. When compared to the parameters resulting from
1987. A general model for single-sludge wastewater treatment
a model with integrated hydraulics, considerable deviations
systems. Water Research 21 (5), 505–515.
may occur. In the end, this can lead to larger reactor sizes in Hocine, S., Pibouleau, L., Azzaro-Pantel, C., Domenech, S., 2008.
case of underprediction of the plant capacity or to plant failure Modelling systems defined by RTD curves. Computers &
in case of overprediction through capacity overload. Chemical Engineering 32 (12), 3112–3120.
1776 water research 44 (2010) 1765–1776

Kjellstrand, R., Mattsson, A., Niklasson, C., Taherzadeh, M.J., 2005. chemical reactions in such systems. Transactions of
Short circuiting in a denitrifying activated sludge tank. Water American Institute of Chemical Engineers 31, 409–458.
Science and Technology 52 (10–11), 79–87. Milbury, W., Pipes, W., Grieves, R., 1965. Compartmentalization of
Koch, G., Kuhni, M., Gujer, W., Siegrist, H., 2000. Calibration and aeration tanks. Journal of the Sanitary Engineering Division
validation of activated sludge model No. 3 for Swiss municipal ASCE 91 (SA3), 45.
wastewater. Water Research 34 (14), 3580–3590. Mills, P.L., Dudukovic, M.P., 1988. Deconvolution of noisy tracer
La Cour Jansen, J.I., Kristensen, G.H., Laursen, K.D., 1992. response data by a linear-filtering method. AIChE Journal 34
Activated-sludge nitrification in temperate climate. Water (10), 1752–1756.
Science and Technology 25 (4–5), 177–184. Nash, J.E., Sutcliffe, J.V., 1970. River flow forecasting through
Lagarias, J.C., Reeds, J.A., Wright, M.H., Wright, P.E., 1998. conceptual models part I – a discussion of principles. Journal
Convergence properties of the Nelder–Mead simplex method in of Hydrology 10 (3), 282.
low dimensions. SIAM Journal of Optimization 9 (1), 112–147. Nauman, E.B., 1969. Residence time distribution theory for
Langergraber, G., Rieger, L., Winkler, S., Alex, J., Wiese, J., unsteady stirred tank reactors. Chemical Engineering Science
Owerdieck, C., Ahnert, M., Simon, J., Maurer, M., 2004. A 24 (9), 1461–1470.
guideline for simulation studies of wastewater treatment Niemi, A.J., 1977. Residence time distributions of variable flow
plants. Water Science and Technology 50 (7), 131–138. processes. International Journal of Applied Radiation and
Leclerc, J.P., Claudel, S., Lintz, H.G., Potier, O., Antoine, B., 2000. Isotopes 28 (10–11), 855–860.
Theoretical interpretation of residence-time distribution Potier, O., Leclerc, J.-P., Pons, M.-N., 2005. Influence of geometrical
measurements in industrial processes. Oil & Gas Science and and operational parameters on the axial dispersion in an
Technology-Revue De L Institut Francais Du Petrole 55 (2), 159–169. aerated channel reactor. Water Research 39 (18), 4454.
Lemoullec, Y., Potier, O., Gentric, C., Leclerc, J.P., 2008. A general Ramirez, F.P., Cortes, M.E., 2004. The determination of residence
correlation to predict axial dispersion coefficients in aerated times in a pilot plant. Nuclear Instruments & Methods in
channel reactors. Water Research 42 (6–7), 1767–1777. Physics Research Section B – Beam Interactions with Materials
Levenspiel, O., 1999. Chemical Reaction Engineering, third ed. and Atoms 213, 144–148.
Wiley & Sons, New York. Thyn, J., Zitny, R., 1996. Problems of residence time distribution
Long, C.A., 1999. Essential Heat Transfer, first ed. Longman, analysis with applications of radiotracers. Journal of
Harlow. Radioanalytical and Nuclear Chemistry 205 (2), 225–233.
Makinia, J., Wells, S.A., 2000. A general model of the activated Tizghadam, M., Dagot, C., Baudu, M., 2008. Wastewater treatment
sludge reactor with dispersive flow – I. Model development in a hybrid activated sludge baffled reactor. Journal of
and parameter estimation. Water Research 34 (16), 3987– Hazardous Materials 154 (1–3), 550–557.
3996. Vrabel, P., van der Lans, R., Cui, Y.Q., Luyben, K., 1999.
Mayr, B., Horvat, P., Moser, A., 1992. Engineering approach to Compartment model approach: mixing in large scale aerated
mixing quantification in bioreactors. Bioprocess Engineering 8 reactors with multiple impellers. Chemical Engineering
(3–4), 137–143. Research & Design 77 (A4), 291–302.
MacMullin, R.B., Weber, M., 1935. The theory of short-circuiting in Werner, T.M., Kadlec, R.H., 2000. Wetland residence time
continuous-flow mixing vessels in series and kinetics of distribution modeling. Ecological Engineering 15 (1–2), 77–90.