water

Article
Empirical Sewer Water Quality Model for Generating
Influent Data for WWTP Modelling
Jeroen Langeveld 1,2, * , Petra Van Daal 2,3 , Remy Schilperoort 1 , Ingmar Nopens 4 ,
Tony Flameling 5 and Stefan Weijers 5
1 Partners4UrbanWater, Javastraat 104a, Nijmegen 6524 MJ, The Netherlands;
remy.schilperoort@urbanwater.nl
2 Section Sanitary Engineering, Department of Water Management, Delft University of Technology,
P.O. Box 5048, Delft 2628 CD, The Netherlands; p.m.m.vandaal-rombouts@tudelft.nl
3 Witteveen+Bos, P.O. Box 233, Deventer 7400 AE, The Netherlands
4 BIOMATH, Department of Mathematical Modelling, Statistics and Bioinformatics, Ghent University,
Coupure Links 653, Gent 9000, Belgium; Ingmar.nopens@ugent.be
5 Waterschap De Dommel, P.O. Box 10.001, Boxtel 5280 DA, The Netherlands;
tflameling@dommel.nl (T.F.); sweijers@dommel.nl (S.W.)
* Correspondence: j.g.langeveld@tudelft.nl; Tel.: + 31-6-1897-6283

Received: 30 March 2017; Accepted: 30 June 2017; Published: 5 July 2017

Abstract: Wastewater treatment plants (WWTP) typically have a service life of several decades.
During this service life, external factors, such as changes in the effluent standards or the loading
of the WWTP may change, requiring WWTP performance to be optimized. WWTP modelling is
widely accepted as a means to assess and optimize WWTP performance. One of the challenges for
WWTP modelling remains the prediction of water quality at the inlet of a WWTP. Recent applications
of water quality sensors have resulted in long time series of WWTP influent quality, containing
valuable information on the response of influent quality to e.g., storm events. This allows the
development of empirical models to predict influent quality. This paper proposes a new approach
for water quality modelling, which uses the measured hydraulic dynamics of the WWTP influent
to derive the influent water quality. The model can also be based on simulated influent hydraulics
as input. Possible applications of the model are filling gaps in time series used as input for WWTP
models or to assess the impact of measures such as real time control (RTC) on the performance of
wastewater systems.

Keywords: sewer system; empirical model; influent generator; influent modelling; gap filling

1. Introduction
Modelling of wastewater treatment plants (WWTPs) using activated sludge models (ASM)
has become a standard in both industry and academia for a range of objectives, amongst others
WWTP design [1,2], operation [3], and control [4,5]. Especially the latter objective, developing
control strategies and assessing the performance of control strategies requires high frequency influent
data [6]. In a comprehensive review, [7] discussed the available approaches for generating influent
data. The approaches range from (1) data driven methods based on creating databases with monitoring
and experimental data and derived models using the data; (2) very simple models based on harmonic
functions; and (3) phenomenological models [8,9].
The data driven methods comprise two different approaches. The first method [10] uses pollutant
release patterns derived from literature to generate dynamic influent data aggregating the punctual
emissions from the database. The second method [11] interpolates available influent data at e.g., a
daily timescale to e.g., hourly dynamics.

Water 2017, 9, 491; doi:10.3390/w9070491 www.mdpi.com/journal/water

Water 2017, 9, 491 2 of 18

The simple models based on harmonic functions are very well suited for the analyses of dry
weather flow (DWF) situations, but less so for wet weather flow (WWF) situations [7].
The phenomenological models are the most detailed influent models, that can give a
phenomenological representation of dynamics of WWTP influent, including diurnal patterns, weekend,
seasonal, and holiday variations as well as rain events [7,8]. Despite being labelled as ‘promising’ [7],
todays phenomenological models cannot adequately reproduce the dynamics in WWTP influent
during wet weather due to a relatively poor representation of the build-up and wash off of urban
pollutants. This is also true for the most recently published influent generator by [12], who use a mix
of statistical and conceptual modelling techniques for synthetic generation of influent time series.
The limitations associated with the use of influent generators are not surprising given the state
of the art knowledge on the physical-chemical, biological and transport processes occurring in sewer
systems [13–17]. Sediment transport especially is not very well understood and not very successfully
reproduced in deterministic sewer models. This is partly due to the fact that it is currently not possible
to get enough data on the initial sewer sediment conditions throughout an entire sewer network.
It is interesting to note that the developers of influent generators, showing many similarities with
simplified or parsimonious sewer models [18,19], are facing the same issues as sewer modelers in the
past, i.e., how to incorporate the contribution of in-sewer stocks during storm events to the outflow of
sewers via either combined sewer overflow (CSO) or WWTP influent.
In order to overcome the limitations of deterministic sewer models, regression models have
been proposed, which are validated against monitoring data. A recent successful example of this
approach is given by [20], who developed an empirical model for storm water total suspended solids
(TSS), event mean concentrations with rainfall depth, and antecedent dry weather period as input
variables. These empirical relations, that are valid at a CSO or storm sewer outfall (SSO), however, are
not suitable for the prediction of WWTP influent quality, as these models do not predict the influent
quality during DWF. For WWTP influent modelling, the empirical model described by [21], relates the
influent concentration to the daily flows. A weak point of this approach is the impossibility to account
for the dynamics during storm events. This is a major drawback, as storm events typically do not last
a full day.
Recent applications of water quality sensors have resulted in the availability of long time series
of WWTP influent quantity and quality [22,23]. These time series contain a lot of information on
the response of influent quality to storm events and the contribution of in-sewer stocks to WWTP
influent [24].
In this study, time series analysis is used to understand the dynamics of WWF related variations
in WWTP influent quality and to relate the variation in influent quality to influent hydraulics.
This allowed the development of an empirical model based on understanding of the underlying
physical processes.
The paper is organized as follows: first, the available data set of WWTP Eindhoven is described.
Second, the model development and calibration method are presented. Next, the calibration results,
model results and the transferability of the concept are presented, discussed, and finally, some foreseen
applications of the model are introduced.

2. Materials and Methods

2.1. System Description: The Dommel River IUWS

The Dommel river is relatively small and sensitive to loadings from CSOs and WWTP effluent.
The river flows through the city of Eindhoven (the Netherlands) from the Belgian border (south)
into the River Meuse (north). The Dommel receives discharges from the 750,000 people equivalent
(PE) WWTP of Eindhoven and from over 200 CSOs in 10 municipalities. In summer time, the base
flow in the river just downstream of the WWTP comprises 50% of WWTP effluent, increasing to 90%
during small storm events. The Dommel River does not yet meet the requirements of the European
Water 2017, 9, 491 3 of 18
Water 2017, 9, 491 3 of 18

Union
oxygen Water
(DO)Framework
depletion, Directive
ammonia(WFD) peaks [25]. The water
and seasonal qualitynutrient
average issues toconcentration
be addressed are dissolved
levels [26,27].
oxygen (DO) depletion, ammonia peaks and seasonal average
Earlier research within the KALLISTO project [18,25] demonstrated that the WWTP effluentnutrient concentration levels [26,27].
is the
Earlier research
main source forwithin the KALLISTO
the toxic ammonia peaks project in [18,25]
the Dommeldemonstrated
river and that thethe
that WWTP effluentpeaks
ammonium is the in
main
the
source for the toxic ammonia peaks in the Dommel river and that the ammonium
WWTP effluent can be significantly reduced by applied integrated real time control (RTC). In [28], peaks in the WWTP
effluent
the use ofcanRTC be significantly
by activating reduced
in-sewerby applied
storage integrated
volume to reducerealand
time control
delay (RTC). In [28],
the hydraulic peak the use
loading
of
ofRTC
the by
WWTPactivating
during in-sewer
storm storage
events volume
has been to shown
reduce and to bedelay the hydraulic
an effective peak Reference
measure. loading of the [29]
WWTP during storm events has been shown to be an effective
introduced a new RTC concept: the smart buffer, which minimizes the peak load to the measure. Reference [29] introduced
biology at a
new
WWTP RTCEindhoven
concept: the bysmart buffer,
applying which
the minimizes the
aforementioned RTC peak load to the
combined withbiology
usingat WWTP
only one Eindhoven
of the three
by applying the aforementioned RTC combined with using only
primary clarifiers (PC) during dry weather (DWF) and using the other two PCs only during one of the three primary clarifiers
storm
(PC) during
events. dry weather (DWF) and using the other two PCs only during storm events.
The
The10 10municipalities
municipalitiescontributing
contributingto tothe
theWWTP
WWTPinfluent
influentare aredivided
dividedover
overthree
threecatchment
catchmentareas areas
that are very different in size and character, each having a separate inflow
that are very different in size and character, each having a separate inflow to the WWTP (see Figure to the WWTP (see Figure 1).
Wastewater from Eindhoven Stad (ES, municipality of Eindhoven)
1). Wastewater from Eindhoven Stad (ES, municipality of Eindhoven) accounts for approximately accounts for approximately 50%
(in
50%practice ranging
(in practice between
ranging between14,000 and and
14,000 17,000 m3 /h)
17,000 m3/h)of of
thethehydraulic
hydrauliccapacity
capacityandandisisdischarged
discharged
directly
directly to the WWTP. The other nine (much smaller) municipalities are each connectedto
to the WWTP. The other nine (much smaller) municipalities are each connected toone
oneof ofthe
the
two
twowastewater
wastewater transport
transport mains,
mains, oneone toto the
thenorth
north(Nuenen/Son
(Nuenen/Sonoror NS,NS, 7 km
7 km in in length)
length) and and
oneone to
to the
the south (Riool Zuid or RZ, 32 km in length), accounting for respectively 7% (3000 m 3 /h) and 43%
south (Riool Zuid or RZ, 32 km in length), accounting for respectively 7% (3000 m /h) and 43% (in 3

(in practice ranging from 14,000 3

practice ranging from 14,000 to to 15,000
15,000 m3m /h)/h)
of theof the hydraulic
hydraulic capacity.
capacity. AnAn elaborate
elaborate description
description of
of the
the studied
studied wastewater
wastewater system
system cancanbe be found
found in [23].
in [23].

Closing
section
catchment area:
ps Son river
Nuenen/Son
ps Nederwetten
Eindhoven Stad
Riool-Zuid ps Gerwen
ps Nuenen

wwtp cs rwzi
Eindhoven ps Eeneind
ps Wintelre ps Mierlo

sludge
Veldhoven Geldrop proc. inst.
ps Aalst
ps Knegsel cs de Meren
Aalst
ps Heeze
ps Waalre
ps Steensel
cs Valkenswaard
ps Eersel ps Sterksel
ps Duizel Valkenswaard
ps Riethoven ps Leende

Bergeijk ps Westerhoven wwtp Eindhoven

ps Weebosch free flow conduits
ps Borkel pressure mains
pumping stations
control stations
Luijksgestel
sludge proc. inst.
N
0 3 6 9 km

Figure1.1.Wastewater
Figure Wastewatersystem
systemof
ofEindhoven
Eindhoven(Left)
(Left)and
andits
itsreceiving
receivingstreams
streamsand
andschematic
schematiclay
layout
outof
of
the wastewater system; (Right) Figure reproduced with permission from
the wastewater system; (Right) Figure reproduced with permission from [23].[23].

2.2. Monitoring Network and Data Validation

2.2. Monitoring Network and Data Validation
At each of the three inflows into the WWTP (locations ‘A’ in Figure 1 at the right) on-line
At each of the three inflows into the WWTP (locations ‘A’ in Figure 1 at the right) on-line
spectroscopy sensors (UV-VIS) have been installed that measure equivalent concentration values of
spectroscopy sensors (UV-VIS) have been installed that measure equivalent concentration values of
wastewater quality parameters: total suspended solids (TSS), chemical oxygen demand (COD), and
wastewater quality parameters: total suspended solids (TSS), chemical oxygen demand (COD), and
filtered COD (CODf), i.e., the dissolved fraction of COD, at an interval of 2 min. In addition, flow is
filtered COD (CODf), i.e., the dissolved fraction of COD, at an interval of 2 min. In addition, flow is
recorded every minute at these locations and ammonium (NH4, using Hach Lange Amtax sensors) at
recorded every minute at these locations and ammonium (NH4 , using Hach Lange Amtax sensors)
the Eindhoven Stad and Riool Zuid catchments. In this study, monitoring data for the year 2012 was
at the Eindhoven Stad and Riool Zuid catchments. In this study, monitoring data for the year 2012
used.
was used.
The monitoring data were validated manually, focusing on obtaining reliable data for calibration
The monitoring data were validated manually, focusing on obtaining reliable data for calibration of
of wet weather flow processes. Figure 2 shows an example of data and their evaluation. After
wet weather flow processes. Figure 2 shows an example of data and their evaluation. After validation,
validation, only 38.5% of the data was considered to have an acceptable quality during the condition
required. This percentage of data perceived ‘good enough’ after validation may seem relatively low.
Water 2017, 9, 491 4 of 18

Water 2017, 9, 491 4 of 18

only 38.5% of the data was considered to have an acceptable quality during the condition required.
During
This earlier research
percentage of dataprojects (2007–2008)
perceived at WWTP
‘good enough’ afterEindhoven
validationonmayUV-VIS
seem sensors, thelow.
relatively percentage
During
of ‘good enough’ data after data validation ranged between 50% and 75%, despite
earlier research projects (2007–2008) at WWTP Eindhoven on UV-VIS sensors, the percentage of ‘goodvery intensive
maintenance
enough’ and data
data after surveillance
validation[23] and between
ranged without 50%restrictions
and 75%,ondespite
the influent conditions.
very intensive WWTP
maintenance
influent has shown
and surveillance [23]toand
be without
a very difficult medium
restrictions on thefor waterconditions.
influent quality monitoring. The dataset
WWTP influent after
has shown
validation comprises approximately 30 storm events with good data for each calibration
to be a very difficult medium for water quality monitoring. The dataset after validation comprises performed.
In the model calibration,
approximately thesewith
30 storm events events,
goodincluding the antecedent
data for each dry day and
calibration performed. Inseveral following
the model dry
calibration,
days, were used.
these events, In Figure
including 2, and all other
the antecedent Figures,
dry day the datafollowing
and several used for calibration is represented
dry days, were by the
used. In Figure 2,
and all other Figures, the data used for calibration is represented by the dark grey bullets, data24
dark grey bullets, data not used for calibration with light grey bullets. An assessment of routine noth
water quality
used for samples
calibration of WWTP
with influent
light grey showed
bullets. that the DWF
An assessment does not
of routine show
24 h watera noticeable seasonal
quality samples of
pattern.
WWTP influent showed that the DWF does not show a noticeable seasonal pattern.

Figure 2. Example of quality evaluation of monitoring data.

2.3. Data
2.3. Data Analysis
Analysis
In earlier
In earlier work
work [16],
[16], aa part
part of
of this
this data
data set
set was
was used
used to
to study
study the
the dynamics
dynamics of
of wastewater
wastewater
composition. Thisresulted
composition. This resulted in well
in well described
described typical
typical diurnal
diurnal patterns
patterns during
during DWF DWF
and and
typical typical
dynamics
dynamics during WWF
during WWF (Figure 3). (Figure 3).

Figure 3. Wet
Figure 3. weather flow
Wet weather flow (WWF)
(WWF) dynamics
dynamics during
during the
the stages
stages of
of aa storm
storm event
event on
on 12
12 June 2008 in
June 2008 in
Eindhoven Stad catchment.
Eindhoven Stad catchment.
 Water2017,
Water 2017,9,9,491
491 55ofof1818

For WWF, it has been observed that the concentration levels of the wastewater show a typical
For WWF,
pattern duringitahas stormbeenevent:
observed thatperiod
a short the concentration
called ‘onset’ levels
of the of storm
the wastewater
event, with showan aincreased
typical
pattern during a storm event: a short period called ‘onset’ of the
concentration level for particulate matter but not for dissolved matter, a longer period calledstorm event, with an increased
concentration
‘dilution’, where level dilution
for particulate
of bothmatter but notand
dissolved for dissolved
particulatematter,mattera longer periodand
takes place, called ‘dilution’,a
‘recovery’,
where
perioddilution of both dissolved
where dissolved and particulate
and particulate matterreturn
matter slowly takes to place,
DWFand ‘recovery’, a period where
levels.
dissolved and particulate matter slowly return to DWF levels.
2.4. Model Development
2.4. Model Development
The general idea behind the model development is that the measured hydraulic influent data,
i.e., flowgeneral
The and water idealevel
behind theinfluent
in the model pumping
development is that
station, canthe measured
be used to makehydraulic influent
a distinction data,
between
i.e.,
theflow
four and waterDWF,
patterns: level onset
in theofinfluent
WWF, pumping station,WWF
dilution during can beand used to make
recovery a distinction
after WWF. Each between
of these
the four patterns: DWF, onset of WWF, dilution during WWF and recovery
patterns is denoted as a system state, during which a certain relation between flow and concentration after WWF. Each of these
patterns is denoted as a system state, during which a certain relation between
level applies. This allows the incorporation of the contribution of in-sewer stocks on top of the mixing flow and concentration
level applies.
process betweenThis allows
wastewater the incorporation of the The
and stormwater. contribution
latter is of in-sewer stocks
a common featureonoftop of the mixing
influent models
process between wastewater and stormwater. The latter is a common feature
applied to simulate both dry and wet periods, while explicitly accounting for the contribution of in- of influent models applied
tosewer
simulate
stocks both dry and wetthe
circumventing periods, while
relatively explicitly
limited accounting
knowledge for the
associated incontribution
sewer processes. of in-sewer
stocksThe circumventing
water quality theaverage
relativelydrylimited
weather knowledge associated
diurnal pattern is thein core
sewer ofprocesses.
the model. As long as the
The water quality average dry weather diurnal pattern
system state is ‘DWF’, the average dry weather diurnal pattern based on monitoring is the core of the model. Asdata longisasused,
the
system
together state
with is the
‘DWF’, the average
measured dry weather
flow data. The averagediurnal drypattern
weather based on monitoring
diurnal pattern has data beenisderived
used,
together
from flow monitoring data by averaging the monitoring data of 10 dry days over 5 min intervalsfrom
with the measured flow data. The average dry weather diurnal pattern has been derived with
flow monitoring
the same timestamp.data by averaging the monitoring data of 10 dry days over 5 min intervals with the
same timestamp.
During wet weather, the model superimposes a number of processes on the DWF pattern for
During
water quality wetto weather,
mimic onset, the model superimposes
dilution, and recovery. a number
The typeofofprocesses
parameter on(NH
the 4DWF
or COD) pattern
andforthe
water quality to mimic onset, dilution, and recovery. The type
type of event (small, medium or large) determine which of these processes is to be of parameter (NH or COD) and
4 applied. The type the
type of event
of event is used(small, medium or large)
as characteristic determine
of a storm event,which
as it wasof these
found processes
that the is to be applied.
relation betweenThe flowtype
and
ofconcentration
event is used levelsas characteristic of a storm event, as it was found that the
differs very much between small, medium, and large storm events. The relation between flow and
concentration levels differs
measured hydraulics, very
in this much
case thebetween
flow andsmall,
watermedium,
level in theandinfluent
large storm events.station,
pumping The measured
are used
hydraulics, in this case the flow and water level in the influent pumping
to determine which of the described processes should be activated in the model, using the scheme ofstation, are used to determine
which
Figureof4.the described processes should be activated in the model, using the scheme of Figure 4.

yes
DWF, process 1

no

yes
DWF, process 1

no

yes
∩
small event, process 6 dilution and restoration

no

yes
∩ medium event, process 2 dilution, process 3 dilution during onset,
process 5 first flush during onset (for COD only) and process 4 recovery

no

yes large event, process 2 dilution, process 3 dilution during onset, process

5 first flush during onset (for COD only) and process 4 recovery

Figure 4. Selection of water quality processes to be superimposed on process 1 during dry weather
Figure 4. Selection of water quality processes to be superimposed on process 1 during dry weather
(DWF), using information on hydraulics. The abbreviation Th in the figure is short for threshold.
(DWF), using information on hydraulics. The abbreviation Th in the figure is short for threshold.
Water 2017, 9, 491 6 of 18

As indicated in Figure 4, two conditions have to be met to change from DWF to WWF. The first is
that the upper limit for dry weather conditions (QDWF , set at the 95th percentile of the flow values
collected during dry weather at a specific timestamp) has to be exceeded, the second that the volume
should exceed a certain threshold (set at 5000 m3 ). The second condition is added to exclude apparent
events in the data caused by interference of the pump operation due to for example to maintenance.
The value of 5000 m3 , equivalent to the volume of 2 h of DWF, shown to be sufficient to filter flow
values exceeding QDWF due to operational issues during DWF.
A small storm event is defined as an event for which the water level in the influent chamber does
not rise above the DWF threshold value (set at 11.30 m AD) and the flow exceeds the 95 percentile
DWF value with less than a threshold set at 4000 m3 /h These events are very small storm events,
where the inflow is less than 0.2 mm/h or 2 m3 /ha). Medium events are defined as events where the
water level in the influent chamber does not exceed the DWF threshold value, but the flow exceeds the
95 percentile DWF value with more than the threshold. These events are typically relatively small, low
intensity storm events, where the inflow is less than the available pumping capacity (which is equal to
an interceptor capacity of 0.7 mm/h or 7 m3 /ha). Large storm events are defined as events during
which not only flow increases, but also the water level in the influent pumping station increases above
the DWF threshold value. This occurs only if the sewer system starts filling during bigger storm events
exceeding the pumping capacity of the WWTP.
The processes applied in the model are:
Process 1 the basic process for all parameters, is the DWF pattern for water quality. It is derived
from high-frequency monitoring data collected during multiple dry weather days, by averaging over
the same time stamps.
Process 2 mimics dilution and is based on the ratio between the actual flow (Qactual ) and the
upper limit for the flow during DWF at that time of the day at the location of the WWTP inlet works
(QDWF ). The wastewater concentration is calculated using Formula (1):

CWWF (t) = CDWF (t) (a1 QDWF (t)/Qactual (t) − a1 + 1) (1)

With CWWF = calculated concentration during wet weather, and CDWF = the concentration during
DWF conditions at that time of the day. The dilution factor a1 (-) is introduced to allow adjustment to
the dilution rate if necessary. A value of 1 for factor a1 indicates that the dilution is exactly inverse to
the increase in flow. A value of a1 smaller than 1 would impose an increase in pollutant loads during
the event, which could be necessary to account for pollutant contributions originating from in-sewer
stocks. A value of a1 larger than 1 would impose a decrease in pollutant loads during the event, which
could be expected for a compound where in-sewer stocks are zero and a part of the pollutants would
be discharged via a CSO. For low dilution ratios, i.e., QDWF (t)/Qactual (t) being close to 1, the factor a1
has a limited influence, for higher dilution ratios, the factor a1 contributes to a larger extent.
Process 3 accounts for dilution during the onset of storm events. Process 3 is described by a
parabolic function, valid for the period between the start of the storm event and the moment of the
maximum dilution. The length of this period, the duration of the onset a3 , is determined during model
calibration. This process is necessary to account for the delayed dilution observed in the monitoring
data, see Figure 5. During the first part of the event on 28 July 2012, the influent flow increased
rapidly to the maximum flow rate of 14,000 m3 /h, while the NH4 concentration gradually reduced to
a minimum of 5 mg NH4 /l. Using Formula (1) during the onset of the storm event, would result in an
overestimation of the dilution. Instead, Formula (3) applies during the onset of the storm event:

CWWF (t) = CDWF (t) × (dilution depth/a2 2 ) × (t − tend onset )2 - dilution depth + 1 (2)

where a2 = duration of onset and dilution depth = the minimal ratio CDWF (t)/CWWF (t) during the
onset of the storm event. Figure 5 shows how the parameters in the parabolic function of Formula (2)
are defined.
Water 2017, 9, 491 7 of 18
Water 2017, 9, 491 7 of 18

Figure 5. Parameters of parabolic function that describes the delayed dilution of the quality
Figure 5. Parameters of parabolic function that describes the delayed dilution of the quality parameters
compared tocompared
parameters the flow. to the flow.

Process 4 reproduces restoration, which describes the gradual return of concentration values to
Process 4 reproduces restoration, which describes the gradual return of concentration values to
DWF values after the storm event. Based on the analysis of the available data set, restoration can be
DWF values after the storm event. Based on the analysis of the available data set, restoration can be
assumed to be a linear process at rate a3 (mg/(L·s)) until the concentration returns to the DWF value.
assumed to be a linear process at rate a3 (mg/(L·s)) until the concentration returns to the DWF value.
During the restoration phase, the concentration is calculated by:
During the restoration phase, the concentration is calculated by:
CWWF(t + 1) = CWWF(t) (1 + a3) dt. (3)
CWWF (t + 1) = CWWF (t) (1 + a3 ) dt. (3)
Process 5 describes a first flush in concentration levels of particulate material (see Figure 3), it is
thus not valid for soluble substances. This initial peak increases the concentrations during the first
Process 5 describes a first flush in concentration levels of particulate material (see Figure 3), it
stage of storm events, before dilution becomes the dominant process. Process 5 is modeled as a
is thus not valid for soluble substances. This initial peak increases the concentrations during the
triangle that causes an instant increase of the COD concentration of a4 mg/l at the onset of the event,
first stage of storm events, before dilution becomes the dominant process. Process 5 is modeled as a
decreasing with a fixed rate a5 (mg/(L·s)).
triangle that causes an instant increase of the COD concentration of a4 mg/l at the onset of the event,
Process 6 regards dilution and restoration for small events. Process 6 describes the concentration
decreasing with a fixed rate a5 (mg/(L·s)).
profile as a fixed-shape triangle, where dilution takes place at a fixed rate a6 (mg/(L·s)) during x h and
Process 6 regards dilution and restoration for small events. Process 6 describes the concentration
recovery at the same rate a6 (mg/L/s) during the next x h. In the case of Eindhoven Stad and Riool
profile as a fixed-shape triangle, where dilution takes place at a fixed rate a6 (mg/(L·s)) during x h and
Zuid a duration of 13 h proved to be a good estimate of the duration of process 6.
recovery at the same rate a6 (mg/L/s) during the next x h. In the case of Eindhoven Stad and Riool
Zuid a duration of 13 h proved to be a good estimate of the duration of process 6.
2.5. Model Calibration
2.5. Model
In thisCalibration
study the differential evolution adaptive metropolis (DREAM) algorithm is the method
[30,31] applied
In this study to calibrate the parameters
the differential of the
evolution empirical
adaptive model to(DREAM)
metropolis find the minimal difference
algorithm is the
between the empirical model output and the monitoring data. The effectiveness of DREAM
method [30,31] applied to calibrate the parameters of the empirical model to find the minimal difference in water
related model
between calibration
the empirical has output
model been demonstrated in manydata.
and the monitoring previous studies, e.g.,of
The effectiveness [32–34].
DREAM in water
related model calibration has been demonstrated in many previous studies, e.g., [32–34]. procedure.
Table 1 shows the model parameters, units and the searching range for the calibration
The threshold
Table 1 shows values
the for selecting
model the type
parameters, of and
units event
thewere derived
searching during
range data
for the analysis procedure.
calibration before the
calibration
The threshold of the model
values for parameters
selecting the and were
type consequently
of event not included
were derived during in theanalysis
data model calibration.
before the
Future users of the model on other catchments may include these
calibration of the model parameters and were consequently not included in the modelparameters as part ofcalibration.
the model
calibration.
Future usersFor of reasons
the model of clarity,
on otherthese parameters
catchments mayare listed here:
include these parameters as part of the model
V Th: threshold value for making distinction between real storm events and irregularities in the
calibration. For reasons of clarity, these parameters are listed here:
DWFVdue to operational issues. In this study set at equivalent of 2 h of DWF.
Th : threshold value for making distinction between real storm events and irregularities in the
DWFQdue Th: threshold value to distinguish medium from small storm events. Set in this study at an
to operational issues. In this study set at equivalent of 2 h of DWF.
equivalent of 0.2 mm/h
QTh : threshold of runoff.
value to distinguish medium from small storm events. Set in this study at an
hTh: threshold value to distinguish large from medium storm events. Set in this study at 0.30 m
equivalent of 0.2 mm/h of runoff.
abovehTh the setpoint of
: threshold the frequency
value controlled
to distinguish pumps.
large from medium storm events. Set in this study at 0.30 m
x: duration of dilution and restoration
above the setpoint of the frequency controlled pumps. for small events: Set in this study at 13 h.
x: duration of dilution and restoration for small events: Set in this study at 13 h.
Water 2017, 9, 491 8 of 18

Table 1. Model parameters, units, and search range. COD = chemical oxygen demand.

Model Parameter Abbreviation Unit Search Range Parameter

dilution factor a1 - 0–2 NH4 , COD
dilution delay factor a2 minutes 0–600 NH4 , COD
recovery factor a3 mg/(L·s) 0–0.01 NH4 , COD
Water 2017, 9, 491 8 of 18
peak first flush concentration a4 mg/l 0–2000 COD
recovery factor first flush a mg/(L·s) 0–10
Table 1. Model parameters, 5units, and search range. COD = chemical oxygen demand.
COD
recovery factor small events a6 mg/(L·s) 0–0.01 NH4 , COD
Model Parameter Abbreviation Unit Search Range Parameter
dilution factor a1 - 0–2 NH4, COD
The calibration isdilution delay factor
performed a2
using 5000 iterations minutes
in DREAM for 0–600
the CODNH 4, COD
model and 2500 for the
recovery factor a3 mg/(L·s) 0–0.01 NH4, COD
NH4 model, as it peak
wasfirst
found from test
flush concentration runs thata4 the cumulative
mg/l density
0–2000 functions
COD of the parameters
do not change (within thefactor
recovery parameter
first flush stability)a5after a mg/(L·s)
few thousand 0–10 iterations.
COD The last 50% of the
recovery factor small events a6 mg/(L·s) 0–0.01 NH4, COD
iterations are used for further analysis: the optimal parameter set and model output are derived, and
the model is run Thewith all these
calibration parameter
is performed usingsets
5000to determine
iterations the 95%
in DREAM confidence
for the COD model intervals
and 2500 for
for the NH4
and COD concentrations.
the NH4 model, as it was found from test runs that the cumulative density functions of the parameters
do not change (within the parameter stability) after a few thousand iterations. The last 50% of the
3. Results iterations are used for further analysis: the optimal parameter set and model output are derived, and
and Discussion
the model is run with all these parameter sets to determine the 95% confidence intervals for the NH4
This section
and CODpresents an overview of the model results, a discussion of their quality and the
concentrations.
transferability of the model concept. The model was first developed, tested and calibrated on data of
3. Results and Discussion
Eindhoven Stad only and the data from the subcatchment Riool Zuid were used to be able to discuss
This section presents an overview of the model results, a discussion of their quality and the
the transferability of the concept.
transferability of the model concept. The model was first developed, tested and calibrated on data of
Eindhoven Stad only and the data from the subcatchment Riool Zuid were used to be able to discuss
3.1. Calibration Results
the transferability of the concept.
The DREAM algorithm was applied with a total of 5000 iterations for the COD model and 2500
3.1. Calibration Results
for the NH4 model. The algorithm uses 2 × n Markov Chains, with n being the number of model
The DREAM algorithm was applied with a total of 5000 iterations for the COD model and 2500
parametersforbeing evaluated. This resulted in 312 iterations for the COD model and 208 iterations for
the NH4 model. The algorithm uses 2 × n Markov Chains, with n being the number of model
the NH4 model. Figure
parameters being 6 shows theThis
evaluated. variation
resulted in model
in 312 parameter
iterations values
for the COD during
model theiterations
and 208 calibration
for process
for the model
the NH for4 model.
NH4 for catchment
Figure 6 shows theEindhoven
variation inStad
modelforparameter
the firstvalues
Markov Chain
during of each iteration.
the calibration
Parametersprocess for the model
concerning medium for NH
and4 for catchment Eindhoven Stad for the first Markov Chain of each
large storm events are denoted in Figure 6 as subscripts M and
iteration. Parameters concerning medium and large storm events are denoted in Figure 6 as
L respectively. The value of each of the model parameters is relatively stable during the calibration
subscripts M and L respectively. The value of each of the model parameters is relatively stable during
process, showing that the
the calibration number
process, showingof that
iterations wasofsufficient
the number to converge.
iterations was sufficient to converge.

Figure 6. Parameter values during model calibration. For each iteration, only the first Markov Chain
Figure 6. Parameter values during model calibration. For each iteration, only the first Markov Chain of
of the parameter set is shown.
the parameter set is shown.
The correlation between the model parameters was found to be limited from Figure 7, showing
a high identifiability of the model parameters.
The correlation between the model parameters was found to be limited from Figure 7, showing a
high identifiability of the model parameters.
Water 2017, 9, 491 9 of 18
Water 2017, 9, 491 9 of 18

Figure 7.
7. Posterior
Posterior distribution
distribution of
of model parameters a11 to a66 for the NH44 model for catchment
model parameters
Eindhoven Stad based on the second 50% of the the iterations.
iterations. The histograms represent the posterior
distribution ofindividual
distribution of individualmodel
model parameters
parameters andand the scatter
the scatter plots plots represent
represent the relationships
the relationships for
for various
various combinations
combinations of parameters.
of parameters.

The model parameter

The model parameter values
values areare shown
shown in in Table
Table 22 for
for the
theNH
NH4 and
and COD
COD model.
model. For
For the
the NH
NH4
4 4
model, there
model, there is
is no
no strong need to
strong need to make
make aa distinction
distinction between
between large and medium
large and storms, as
medium storms, as the
the model
model
parameters are
parameters are rather
rather similar.
similar. ForFor COD,
COD, however, the model
however, the model parameters
parameters differ
differ strongly
strongly for
for large
large and
and
medium storm events. A value of 1 of the dilution factor would mean no
medium storm events. A value of 1 of the dilution factor would mean no contribution of in-sewer contribution of in-sewer
stocks and
stocks and only
only perfect
perfect dilution, values below
dilution, values below 11 would
would indicate
indicate aa contribution
contribution of of in-sewer
in-sewer stocks.
stocks. The
The
values for the dilution factor for NH 4 are just below 1, indicating that for NH4 the contribution of in
values for the dilution factor for NH4 are just below 1, indicating that for NH4 the contribution of
sewer
in stocks
sewer is relatively
stocks is relatively limited.
limited.ForFor
thethe
COD, however,
COD, however, thethevalues of aof
values 1 are quite low. A value for
a1 are quite low. A value
a 1 of approximately 0.5 leads to CWWF = 0.57 × CDWF based on Formula (1) and the maximum ratio
for a1 of approximately 0.5 leads to CWWF = 0.57 × CDWF based on Formula (1) and the maximum
between
ratio Qactual Q
between and QDWF (15,000/2100 = 7.1). This means that at a flow of 15,000 m3/h, the concentration 3
actual and QDWF (15,000/2100 = 7.1). This means that at a flow of 15,000 m /h, the
in the WWTP in
concentration influent
the WWTP is stillinfluent
0.57 × Cis DWF. Consequently, the influent load at that moment will be
still 0.57 × CDWF . Consequently, the influent load at that
approximately 4 (7.1 × 0.57) times
moment will be approximately 4 (7.1 × 0.57) the DWFtimes
load.the
This peak
DWF loadThis
load. factor
peakwas observed
load regularly
factor was observed in
monitoring data for this catchment [23].
regularly in monitoring data for this catchment [23].

Table 2.
2. Model
Model parameters
parameters values
values for
for NH
NH4 and COD model for Eindhoven Stad Catchment. Subscript
Table 4 and COD model for Eindhoven Stad Catchment. Subscript
M and L in model parameters denote medium and large storm events.
M and L in model parameters denote medium and large storm events.
Model Parameter Abbreviation NH4 Model COD Model
Model Parameter Abbreviation NH4 Model COD Model
dilution factor, large storms a1,L 0.95 (-) 0.63 (-)
dilution
dilution factor,
delay largelarge
factor, storms storms a1,L
a2,L 0.95
123 (-)
(min) 0.63(min)
342 (-)
dilution delay factor, large storms a2,L 123 (min) 342 (min)
dilution factor, medium storms a1,M 0.82 (-) 0.47 (-)
dilution factor, medium storms a1,M 0.82 (-) 0.47 (-)
dilution
dilution delay factor,medium
delay factor, medium storms
storms a2,M
a2,M 115
115 (min)
(min) 589 (min)
589 (min)
recoveryfactor
recovery factor a3 a3 0.00025
0.00025 (mg/(L·s))
(mg/(L ·s)) 0.00014(mg/(L
0.00014 (mg/(L·s))
·s))
peak first
peak firstflush
flushconcentration
concentration a4 a4 Not
Notapplicable
applicable 48
48(mg
(mgCOD/(L ·s))
COD/(L·s))
recovery factor
recovery factorfirst
firstflush
flush a5 a5 Not
Notapplicable
applicable 0.19
0.19(mg
(mgCOD/(L ·s))
COD/(L·s))
recovery factor small events a6 0.00059 (mg/(L·s)) 0.00017 (mg/(L·s))
recovery factor small events a6 0.00059 (mg/(L·s)) 0.00017 (mg/(L·s))
Water 2017, 9,
Water 2017, 9, 491
491 10
10 of
of 18
18
Water 2017, 9, 491 10 of 18
3.2. Model Results
3.2. Model
3.2. Model Results
Results
Figure 8 shows the resulting predicted water quality and the measured water quality for NH4 in
Figureinfluent
the WWTP
Figure 88 shows
showsfor
theEindhoven
the resulting predicted
resulting predicted
Stad. waterquality
water qualityand
andthe
the measured
measuredwater
waterquality
qualityfor
forNH
NH4 in
in
4
the WWTP influent for Eindhoven
the WWTP influent for Eindhoven Stad. Stad.

Figure 8. Model vs. monitoring data: NH4 concentration in the wastewater treatment plant (WWTP)
Figure 8. Model
Figure
influent. Modelvs.vs.
monitoring data:data:
monitoring NH4 concentration in the wastewater
NH4 concentration treatment
in the wastewater plant (WWTP)
treatment plant
(WWTP)
influent. influent.
The results show that the dynamics in the model (solid black line) and the monitoring data (grey
dots)TheThe
show results
results show that
show
an overall that the
good the dynamicsinin
dynamics
agreement interms
the model
the model (solid black
(solid
of dynamics black line) and
and line)
values and the monitoring
the
during monitoring
wet weather data
data (grey
(grey
during
dots)
dots)
the show
show
large an overall
an
events overall
of 28 and good
good agreement
29agreement
July. In thein in terms of
terms
monitoring of dynamics
dynamics
data, dilution and values
and values during
starts aduring
little bit wet
wet weather
weather
earlier during
thanduring
in the
the large
the large
model. events
eventsthe
During of 28 and
of medium
28 and 29 29 July. In
July.on
event the
In31 the monitoring
monitoring
July, data,
the modeldata, dilution
fit isdilution starts a little
starts aDuring
less satisfying. bit earlier
little bitdry than
earlier in (25
than
weather the
in
model.
the model.
July During the
During
till 28 July), medium
thethe medium
daily event on
event
variation 31
on 31
in the July, the model fit
July, the concentration
measured is less
model fit is less satisfying.
of satisfying. During
NH4 is represented dry
During dry weather
wellweather (25
by the
JulyJuly
(25
model. tillThe
28 28
till July), thethe
July),
remaining daily
dailyvariation
variation
differences inin
are the
due themeasured
the fact concentration
tomeasured the DWF of
concentration
that ofNH
NH44isisrepresented
dynamics the modelwell
represented
in well by the
by
represent the
model.
model.
the The remaining
The remaining
average differences are
differences levels
DWF concentration due
are dueand to
to the the
thefact fact
DWF that
thatvaries the
the DWF DWF dynamics
per dynamics
day. The in root in the
themean model
modelsquared represent
represent the
error
the average
average
(RMSE) DWF
for DWF
the NHconcentration
concentration levelslevels
4 model based andthe
on and
the DWF
data the DWF
varies
used varies perThe
percalibration
in the day. day. isThe
root mean
6.3 rootsquared
mg NHmean squared
error
4/l, or nearly error
(RMSE)
16%
(RMSE)
for the NH tofor the NHbased
4 model based on the data used in the calibration is 6.3 mg NH4/l, or nearly 16%
related 4 model
the mean DWF on the data used
concentration. in the9calibration
Figure shows the is 6.3 mg NHcumulative
normalized 4 /l, or nearly 16% related
density function to
related
the mean
(CDF) to
of thethe
DWF mean
model DWF concentration.
concentration.
results andFigure
the monitoringFigurethe
9 shows 9 shows
data for NHthe4normalized
normalized andcumulative
COD. Both cumulative
density
the RMSE density
function function
and (CDF)
the CDF of
(CDF)
the
are model
based of theon model
results and
the entire results
the and used
the monitoring
monitoring
dataset datamodelling,
for for NH dataandfor COD.
NHincludes
4 and COD. Both the RMSE and the CDF
Both theallRMSE
stormand the with
CDF are based
4 which events sufficient
are
data based
on thequality on the
entire dataset entire
and theused dataset
dry for
days used
modelling,for
preceding modelling,
whichand includes which
following alltheincludes
storm all
events
storm storm
withAs
event. events
sufficient with
expected, data sufficient
thequality
high
datathe
and quality
dry days
concentrations, and the dry
which occurdays
preceding andpreceding
duringfollowing
DWF, are and
thenot following
storm event.
captured the Asstorm
very wellevent.
expected,
due the toAs expected,
high
the the high
concentrations,
daily variation in
concentrations,
which occur which
during DWF,occur
are during
not DWF,
captured are
very not
well captured
due to
DWF concentrations. At lower concentration levels, the agreement between model and monitoringthe very
daily well due
variation toin the
DWF daily variation
concentrations. in
DWF
At lower
data concentrations. At lower concentration levels, the agreement
concentration levels, the agreement between model and monitoring data is reasonable.
is reasonable. between model and monitoring
data is reasonable.

Figure 9. CDFs
Figure 9. CDFs of
of model
model results
results and
and measurements
measurements for NH44 and
for NH and COD
COD in
in WWTP
WWTP influent
influent for
for the
the
Figure 9. CDFs
Eindhoven Stad of model
(ES) results and measurements for NH4 and COD in WWTP influent for the
catchment.
Eindhoven Stad (ES) catchment.
Eindhoven Stad (ES) catchment.
Water 2017, 9, 491 11 of 18
Water 2017, 9, 491 11 of 18

The COD
The CODmodelmodelalso alsoincorporates
incorporates process
process 55describing
describing thethe peak
peak first
firstflush
flushconcentration
concentrationthat that
wasobserved
was observed in inthethemonitoring
monitoringdata,data,seeseeFigure
Figure3.3. The
The model
model results
results for
for the
the COD
COD model
model forfor the
the
Eindhoven Stad catchment are shown in Figure 10. As expected, the model
Eindhoven Stad catchment are shown in Figure 10. As expected, the model fit is not as good as the fit is not as good as the
modelfitfitforfor
model NH NH 4, which
4 , which is partly
is partly due toduetheto the difference
difference in theofquality
in the quality of the monitoring
the monitoring data,
data, illustrated
illustrated
by the outliersby the outliers
in the in the monitoring
monitoring data showndata shown10inand
in Figure Figure 10 due
partly and to
partly duethat
the fact to the fact that
modelling
modelling suspended
suspended solids is much solids is much
more morethan
difficult difficult than modelling
modelling of solutes.ofThe
solutes.
RMSE TheforRMSE
the CODfor the COD
model
modelon
based based on monitoring
monitoring data used data
in used in the calibration
the calibration is 109 mg is 109 mg COD/l,
COD/l, equivalent
equivalent with
with 18% of18% of the
the mean
meanconcentration.
DWF DWF concentration. This RMSE
This RMSE for COD forisCOD is in relative
in relative terms comparable
terms comparable to the RMSE
to the RMSE for NHfor4 . NH4.

Figure 10. Model vs. monitoring data: CODtotal in the WWTP influent for Eindhoven Stad. The small
Figure 10. Model vs. monitoring data: CODtotal in the WWTP influent for Eindhoven Stad. The small
peaks in influent concentration (typically referred to as ‘first flush’) at the beginning of storm events
peaks in influent concentration (typically referred to as ‘first flush’) at the beginning of storm events on
on 4 June, 6 June and 8 June are adequately represented by the model.
4 June, 6 June and 8 June are adequately represented by the model.

The influent model has been developed to be used to deliver input for WWTP models that is
The influenttomodel
reliable enough assesshas
WWTPbeenperformance
developed to beto
and used to deliver
assess input
the impact for WWTPsuch
of measures, models that
as real is
time
reliable enough to assess WWTP performance and to assess the impact of measures,
control (RTC) in wastewater systems. An earlier version of this influent model presented has alreadysuch as real
time
beencontrol
used for(RTC) in wastewater
this purpose systems. An
in the KALLISTO earlieratversion
project of thisDe
water board influent
Dommel model presented
[26,28]. has
The current
already
version been used for
is applied this assessment
in the purpose in of thethe
KALLISTO
performanceproject at water
of the smartboard
bufferDeconcept
Dommel at [26,28].
WWTP
The current version is applied in the assessment of the performance of the smart
Eindhoven [29]. The values for the RMSE for NH4 just meet the quality requirements for WWTP buffer concept at
WWTP
influentEindhoven
data derived[29].
byThe
[35],values
while for
for the
COD RMSE
they for NHmeet
easily 4 justthese
meetrequirements.
the quality requirements for
This shows that
WWTP influent data derived by [35], while for COD they easily meet
the influent model is sufficiently good for the described modelling purposes. these requirements. This shows
that the influent model is sufficiently good for the described modelling purposes.
3.3. Transferability of the Concept
3.3. Transferability of the Concept
The structure of the influent model was developed for the catchment Eindhoven Stad (ES) only.
The structure of the influent model was developed for the catchment Eindhoven Stad (ES) only.
The data from the catchment Riool Zuid (RZ) were used to verify the concept, using the same routines
The data from the catchment Riool Zuid (RZ) were used to verify the concept, using the same routines
for calibration. With respect to the transferability, it has to be noted that the subcatchments ES and
for calibration. With respect to the transferability, it has to be noted that the subcatchments ES and RZ
RZ are independent catchments, allowing the transferability of the concept to be tested.
are independent catchments, allowing the transferability of the concept to be tested.
The model parameters are shown in Table 3. For catchment RZ, the model parameters for
The model parameters are shown in Table 3. For catchment RZ, the model parameters for medium
medium and large storms are very similar for both the NH4 and COD model, showing that this
and large storms are very similar for both the NH4 and COD model, showing that this distinction
distinction between large and medium events is not necessary for this catchment. The model
between large and medium events is not necessary for this catchment. The model parameter values for
parameter values for the dilution factor a1 for the RZ catchment show a strong similarity with the
the dilution factor a1 for the RZ catchment show a strong similarity with the model parameters for the
model parameters for the ES catchment. For NH4, the dilution during the storm event, calculated by
ES catchment. For NH4 , the dilution during the storm event, calculated by Formula (1), remains nearly
Formula (1), remains nearly reciprocal to the increase in influent flow during the storm event. This
reciprocal to the increase in influent flow during the storm event. This indicates that during the storm
indicates that during the storm event, nearly all nitrogen in the WWTP influent stems from the
event, nearly all nitrogen in the WWTP influent stems from the wastewater, with only very limited
wastewater, with only very limited contributions from the rainfall runoff and the in-sewer stocks.
contributions from the rainfall runoff and the in-sewer stocks.
Water 2017, 9, 491 12 of 18
Water 2017, 9, 491 12 of 18
Table 3. Model parameters values for NH4 and COD model for Riool Zuid Catchment. Subscript M
and L in model parameters denote medium and large storm events.
Table 3. Model parameters values for NH4 and COD model for Riool Zuid Catchment. Subscript M
and LModel
in model parameters denote Abbreviation
Parameter medium and large stormNH
events.
4 Model COD Model
dilution factor, large storms a1,L 0.96 (-) 0.49 (-)
Model
dilution delay Parameter
factor, large storms Abbreviation
a2,L NH4(min)
373 Model COD
590Model
(min)
dilution factor,
dilution medium
factor, storms
large storms a1,Ma1,L 0.98
0.96(-)
(-) 0.49
0.49 (-)(-)
dilutiondelay
dilution delay factor,
factor, large
mediumstorms a2,L 373 (min) 590 (min)
dilution factor, medium storms a2,Ma1,M 4270.98
(min)
(-) 548 (-)
0.49 (min)
storms
dilution delay factor, medium storms a2,M 427 (min) 548 (min)
recovery
recovery factor
factor a3 a 0.00033
0.00033(mg/(L·s))
(mg/(L·s)) 0.00034
0.00034 (mg/(L·s))
(mg/(L ·s))
3
peak first
peak flush
first flushconcentration
concentration a 4 a4 Not
Notapplicable
applicable 6060(mg
(mg COD/(L·s))
COD/(L ·s))
recovery
recovery factorfirst
factor firstflush
flush a5 a5 Notapplicable
Not applicable 0.06
0.06(mg(mgCOD/(L ·s))
COD/(L·s))
recovery factor small
recovery factor small events events a6 a6 0.00027 (mg/(L·s))
0.00027 (mg/(L·s)) 0.00002 (mg/(L
0.00002 (mg/(L·s))· s))

For the COD, the dilution factor of a11 of of 0.49

0.49 results
results in
in COD
COD concentration
concentration levels during the high
flow period of storm events of between 250 and 300 300 mgmg COD/L
COD/L and, and, as
as aa consequence,
consequence, high influent
peak loads.
loads.This
Thisadditional
additional load
load arriving
arriving viainfluent
via the the influent at the WWTP
at the WWTP during aduring a storm
storm event event
originates
originates mainly from the in-sewer stocks [15], given the fairly low COD concentration
mainly from the in-sewer stocks [15], given the fairly low COD concentration in Dutch stormwater of in Dutch
stormwater
61 mg COD/L of 61 mg COD/L [36].
[36].
The overall
overallmodel
model performance
performance for RZ
for the thecatchment,
RZ catchment,expressedexpressed
in terms in terms is
of RMSE, ofcomparable
RMSE, is
comparable with the performance for the ES catchment. The RMSE amounts
with the performance for the ES catchment. The RMSE amounts to 8.9 mg NH4 /L, equivalent to 8.9 mg NH 4/L,
with
equivalent with 22%
22% of the mean DWF ofvalue
the mean DWF
for the NHvalue
4 model forand
the to
NH 1264 model and to 126
mg COD/L, mg COD/L,
equivalent with equivalent
25% of the
with 25% of the mean DWF value for the COD model. As for Eindhoven Stad,
mean DWF value for the COD model. As for Eindhoven Stad, the cumulative density function the cumulative density
for
function for model results and monitoring data show reasonable agreement
model results and monitoring data show reasonable agreement for both NH4 and COD, see Figurefor both NH 4 and COD,
11,
see
withFigure 11, with
once again theonce again
biggest the biggest
differences differences
being being in
in the higher the higher
(DWF) (DWF) concentration
concentration ranges. ranges.

Figure 11. CDFs

CDFs of
ofmodel
modelresults
resultsand
and measurements
measurements forfor
NHNH 4 and
4 and CODCOD in WWTP
in WWTP influent
influent for
for the the
Riool
Zuid (RZ)
Riool Zuidcatchment.
(RZ) catchment.

As
As there
there was
was no need to
no need to change
change thethe model
model structure,
structure, it
it was
was concluded
concluded that
that the
the model
model isis in
in
principle transferable to other catchments, provided that the dynamics during WWF
principle transferable to other catchments, provided that the dynamics during WWF show similar show similar
patterns
patterns as described in
as described Figure 3.
in Figure 3. Literature
Literature confirms
confirms these
these patterns
patterns to
to be
be fairly
fairly general
general [37,38],
[37,38], AA
number of distinct phases in the influent pollutograph are defined during storm
number of distinct phases in the influent pollutograph are defined during storm events: events:
Phase
Phase 1)1) increase
increaseofof flow
flow rate
rate and
and subsequently
subsequently anan increase
increase of the
of the loadload arriving
arriving at the
at the WWTP
WWTP duedue to
to the
the ‘push’
‘push’ of wastewater
of wastewater with concentration
with DWF DWF concentration levels.
levels. This phaseThis phasedistinct
is more is more thedistinct the more
more wastewater
wastewater is stored downstream in either large interceptor
is stored downstream in either large interceptor sewers or rising mains.sewers or rising mains.
Phase
Phase 2) increasedconcentration
2) increased concentrationof
ofsuspended
suspendedsolids
solids as
as eroded
eroded sewer
sewer sediments
sediments start
start to
to arrive
arrive at
at the
the
WWTP. These sediments are usually transported with a velocity lower than the fluid
WWTP. These sediments are usually transported with a velocity lower than the fluid velocity [14]. velocity [14].
Phase
Phase 3) arrivalofofdiluted
3) arrival dilutedwastewater
wastewateratatthetheWWTP.
WWTP.
Phase 4) return to DWF equilibrium. Equilibrium for dissolved compounds will be reached as soon
as all remaining storm runoff has been transported (pumped) towards the WWTP. Reaching
Water 2017, 9, 491 13 of 18

Phase 4) return to DWF equilibrium. Equilibrium for dissolved compounds will be reached as soon as
all remaining storm runoff has been transported (pumped) towards the WWTP. Reaching equilibrium
for suspended solids may last longer since it takes time before all depressions within the sewer system
are filled again with sediment.
Phases 1 and 2 are both part of the onset of the storm event, phase 3 is similar to the dilution
stage, while phase 4 relates closely to the stage of recovery after the storm events. Despite the need of
further research on the transferability, the similarities in system dynamics strongly indicate a wider
applicability of the model then just for Eindhoven Stad and Riool Zuid.
The consistency in the dilution factors for NH4 and COD for Eindhoven Stad and Riool Zuid
(with dilution factors for NH4 just a little smaller than 1, indicating a small contribution of in-sewer
stocks and dilution factors for COD around 0.5, indicating a large contribution of in-sewer stocks)
demonstrate that the empirical model is able to capture the contribution of in-sewer stocks during
the dilution phase of the event adequately. This is an important benefit of the model compared with
influent generators reviewed by [7]. The differences in model parameters related to the first flush
and recovery after the event seem to be linked to the differences in lay out of the sewer system.
Future research is necessary to further elaborate on the relation between parameter values and physical
characteristics of the catchment. The amount of in-sewer storage relative to the pumping capacity of
the WWTP will likely be related to the length of the recovery period, as these characteristics determine
the emptying time of the sewer system, which will be related to the length of the recovery.
The differences in performance, although relatively small, can be attributed to the different
characteristics of the RZ catchment and possibly also to the limited quality of the sensor data. This is
illustrated by Figures 12 and 13. In Figure 12, the results for NH4 are shown for the RZ catchment
for the week after 16 July and in Figure 13 for the first two weeks of December 2012. During the
events of 17 July and 21 July, the monitoring data show two distinct dilution phases during the events.
One dilution phase occurs at the beginning of the storm event and one at the end. This phenomenon
does not occur during the storm events of 2 and 4 December, but to a lesser extent on 10 December.
The Riool Zuid catchment (the light grey catchments in Figure 1, left) has a very different structure
compared with ‘Eindhoven Stad’ (the dark colored central catchment in Figure 1, left). While the ES
catchment is a gravity system draining to one central point, the RZ catchment is 32 km long with a
number of subcatchments of various sizes at different distances. It is assumed that the double dip
is caused by a difference in transport times for two main areas in this catchment, which causes the
concentrations to drop for a second time during a storm event. The double dilution dip is of course
driven by rainfall and as a result, the spatial variation in the rainfall is the main explanatory factor for
the differences per event.
In the model, this effect could be mimicked by dividing the Riool Zuid catchment into two
catchment basins, with a cut at pumping station Aalst, see Figure 1 (left), and to add the transport time
in the transport sewer to one of the basins. However, as the error made in the model results is relatively
small due to the low influent flows, this adjustment was not considered necessary at this moment.
The base line of the model consists of the mean DWF concentration. Data analysis of available 24 h
composite samples of the WWTP influent showed that there was no seasonal trend for NH4 and COD.
Despite the absence of a seasonal trend, the mean DWF concentration varied during the monitoring
period. Due to a lack of sampling data, it could not be determined whether changes in this DWF
concentration level were due to real changes or due to e.g., a temporary drift of the sensor. Differences
in the main DWF concentration levels were observed for NH4 , e.g., compare the concentration levels on
16 July in Figure 12 (around 40 mg/L) and 1 December in Figure 13 (around 55 mg/L). This difference
is even bigger for COD, see Figure 14. The mean DWF, derived from monitoring data, is nearly 500 mg
COD/L, while the monitoring data on 2 June are nearly 700 mg COD/L. In this study, the model has
been calibrated using the mean DWF derived from all available monitoring data during DWF.
Water 2017, 9, 491 14 of 18
Water
Water2017,
Water 2017,9,
2017, 9,491
9, 491
491 14
14of
14 of18
of 18
18

Figure
Figure 12.
12.
Figure
Figure 12.
12. Model
Model
Model vs.
Modelvs.
vs. monitoring
vs.monitoring
monitoring data:
monitoringdata:
data:NH
data: NH4444concentration
NH
NH concentrationin
concentration
concentration inthe
in
in theWWTP
the
the WWTPinfluent
WWTP
WWTP influentcatchment
influent
influent catchmentRiool-Zuid
catchment
catchment Riool-Zuid
Riool-Zuid
Riool-Zuid
between
between 16 July and 23 July. ‘Double’ dilution dips occur at 17 July around noon and 18 Julyaround
between
between1616
16 July
July
July and
and
and 2323
23 July.
July.
July. ‘Double’
‘Double’
‘Double’ dilution
dilution
dilution dips
dips
dips occur
occur at
at 17
17 July
July
July around
around
around noon
noon
noon and
and
and 18
18
18 July
July
July around
around
around
midnight
midnight
midnight and
and
andonon
on 21
21
21 July
July
July at
at
at 2:00
2:00
2:00 a.m.
a.m.
a.m. and
and
and 12:00
12:00
12:00
midnight and on 21 July at 2:00 a.m. and 12:00 a.m. a.m.
a.m.
a.m.

Figure
Figure 13.
13.
Figure
Figure 13.
13. Model
Model
Model vs.
Modelvs.
vs. monitoring
vs.monitoring
monitoring data:
monitoringdata:
data: NH
NH4444concentration
data:NH
NH concentrationin
concentration
concentration inthe
in
in theWWTP
the
the WWTPinfluent
WWTP
WWTP influentcatchment
influent
influent catchmentRiool-Zuid
catchment
catchment Riool-Zuid
Riool-Zuid
Riool-Zuid
between
between
between1 1
1 December
December
December and
and
and 15
15
15 December.
December.
December.
between 1 December and 15 December.

Figure
Figure 14.
14.
Figure
Figure 14.
14. Model
Model
Model
Model vs.
vs.
vs. monitoring
vs.monitoring
monitoring data:
monitoringdata:
data: COD
data:COD
COD concentration
concentrationin
CODconcentration inWWTP
in WWTPinfluent
WWTP
WWTP influentcatchment
influent catchmentRiool-Zuid.
catchment
catchment Riool-Zuid.
Riool-Zuid.
Riool-Zuid.
Water 2017, 9, 491 15 of 18

Finally, the results shown in this paper are based on NH4 and COD only, where NH4 is
representative for solutes predominantly derived from wastewater and COD for parameters for
which the in-sewer stocks, i.e., sediment and biofilm, contribute significantly to the pollutant load in
the influent during storm events [24]. For WWTP modelling other parameters, such as phosphorous,
also need to be taken into account. Total phosphorous is known to exert similar behavior as the COD
and consequently, the COD model may be used to recalculate the total phosphorous concentration.
Ortho-phosphate, on the other hand, is dissolved and behaves in the sewer like NH4 [37] and
consequently, the NH4 model may be used.

3.4. Applications: Influent Generation, Surveillance of Monitoring Equipment and Gap Filling
The influent model as described in this paper generates influent water quality dynamics using
measured influent hydraulics as input. This application was used in [29] to evaluate the performance
of integrated RTC for urban wastewater system Eindhoven.
Another application of the influent model is described in [28], where (an earlier version of)
the influent model was implemented in the WEST software [28] to generate influent water quality
dynamics based on the simulated influent water quantity by the sewer sub model. In this application,
a representative DWF curve based on monitoring data was used [28]. It would also have been possible
to use harmonic functions for DWF and to use the influent model to complement the time series
with WWF dynamics. In other words, the influent model can be applied on measured or simulated
hydraulics and may also be applied in combination with the harmonic functions described in the
introduction. Moreover, as the empirical influent model is developed to adequately mimic WWF
dynamics, it might be included in the phenomenological influent pollutant disturbance scenario
generator [9], which is the latest version of the phenomenological model developed by [8], replacing
the relatively weak sewer model module of this model.
The influent model can also be used for surveillance of monitoring equipment at the inlet of the
WWTP. e.g., earlier research has shown that the hydraulic monitoring data at WWTP Eindhoven is
very reliable with over 99% good data [23]. Running the influent model continuously on measured
hydraulic data and comparing raw monitoring data of the influent water quality with the simulated
influent water quality data would allow easy detection of anomalies in the monitoring data. This
could be used to alarm operators to check the monitoring equipment. Early detection of problems with
sensors will possibly result in a higher yield of good quality data.
A final application of the influent model discussed in this paper is gap filling in time series [39].
Figure 15 shows an example of the application of gap filling. The dots show the measured NH4
concentration in the influent of WWTP Eindhoven. The dark grey dots are data that could be used e.g.,
for assessing the performance of the influent model, the light grey dots show data that are rejected
during the validation. The black line shows the simulated concentrations by the influent model.
By applying gap filling a continuous time series is generated, the final time series is composed of the
dark grey dots and where these are missing data are filled in using the influent model results.
For some of the potential applications of the model, such as gap filling or data validation, it is
advisable to regularly check the absolute value of the sensor data during DWF to ascertain whether
changes in the DWF concentration level are due to real changes or due to a sensor drift. In case these
are due to e.g., seasonal variation in the influent concentration levels during DWF, this should be
accounted for when applying the influent model. The routine 24-h samples typically available for
WWTP influent may be used to check for a seasonal pattern.
Water 2017, 9, 491 16 of 18
Water 2017, 9, 491 16 of 18

Figure Gap
Figure 15. Gap filling
filling withwith
the the influent
influent model.
model. ExampleExample
for NH4 for NH4 ininfluent
in WWTP WWTPcatchment
influent
catchment
Eindhoven.Eindhoven.

4. Conclusions
4. Conclusions and
and Outlook
Outlook
Modelling of
Modelling of influent
influent quality
quality is is an
an increasingly
increasingly important
important tool tool toto enable
enable WWTPWWTP models models to to
optimize the
optimize the performance
performance of of WWTPs
WWTPs during during wet
wet weather.
weather. The The main
main issue
issue in
in modelling
modelling wastewater
wastewater
quality during
quality during storm
storm events
events is is to
to account
account forfor in-sewer
in-sewer stocks,
stocks, which
which have
have aa varying
varying contribution
contribution to to
the wastewater
the wastewater quality.
quality. Neither traditional
traditional sewer water water quality
quality models
models nor nor thethe available
available influent
influent
generators are
generators are capable
capable of of adequately
adequately addressing
addressing this this issue.
issue. TheThe proposed
proposed empirical
empirical modelmodel is is based
based
on aa detailed
on detailed study
study of of the
the observed
observed water water quality
quality andand predicts
predicts itit by
by combining
combining aa number
number of of actual
actual
processessuch
processes suchas asDWF,
DWF,dilution,
dilution,restoration,
restoration,and and first
first flush.
flush. Overall,
Overall, thethe model
model showsshows that
that it isitable
is able
to
to reasonably predict NH concentration, which is a solute substance, as
reasonably predict NH4 concentration, which is a solute substance, as well as to reasonably predict
4 well as to reasonably predict
COD concentration,
COD concentration, which
whichis is to
toaalarge
largeextent
extentassociated
associatedwith withparticles.
particles.
The model
The model structure
structure was was demonstrated
demonstrated to to bebe transferable
transferable to to aa catchment
catchment with with different
different
characteristics. Eindhoven Stad is a large catchment with the WWTP located near the center, Riool Riool
characteristics. Eindhoven Stad is a large catchment with the WWTP located near the center, Zuid
Zuid comprises
comprises of ainterceptor
of a long long interceptor sewer,sewer,
which which
drains drains the wastewater
the wastewater frommunicipalities
from seven seven municipalities with a
with aofrange
range of catchment
catchment sizes. Due sizes. Due to variation
to spatial spatial variation of the and
of the rainfall rainfall and variation
variation in travelin travel
times, times,
‘double
‘double dilution’
dilution’ dips maydips occur mayin theoccur in thecoming
influent influentfrom comingRioolfrom Zuid.Riool
DespiteZuid. Despite
these thesedynamics
different different
dynamics
and and characteristics,
characteristics, there was no there
needwas no need
to adjust to adjust
the model the model
structure of thestructure
empiricalofinfluent
the empirical
model.
influent model. Future research in catchments in sewer systems with less
Future research in catchments in sewer systems with less in-sewer storage volume is needed to further in-sewer storage volume is
needed the
explore to further explore of
transferability thethetransferability
model concept. of the model concept.
The model
The modelconcept
conceptisisused usedininongoing
ongoingresearch
researchtototest testthethe performance
performance of of
RTC RTCof of smart
smart buffers
buffers at
at WWTP Eindhoven [28]. The model concept could also be used to fill
WWTP Eindhoven [28]. The model concept could also be used to fill gaps in time series for influent gaps in time series for influent
water quality
water quality and
and be be used
used for for advanced
advanced data data validation
validation to to detect
detect outliers
outliers andand drift
drift of
of water
water quality
quality
sensors, as
sensors, as these
these sensors
sensors arearestill
stillvery
veryvulnerable
vulnerableand anddatadataquality
qualitycontrol
controlremains
remainsaadifficult
difficultissue.
issue.

Acknowledgments: The
Acknowledgments: project
The is funded
project is byfunded
KRW-innovation subsidy from AgentschapNL
by KRW-innovation subsidy from (www.agentschapnl.
AgentschapNL
nl). The Dutch Foundation
(www.agentschapnl.nl). forDutch
The Applied Water Research
Foundation (STOWA)
for Applied WaterisResearch
engaged (STOWA)
in the project to communicate
is engaged and
in the project
publish the research results.
to communicate and publish the research results.
Author Contributions: Jeroen Langeveld and Remy Schilperoort developed the influent model, Petra Van Daal
Author Contributions:
improved Jeroen Langeveld
the model structure and performed and Remy Schilperoort
the model developed
calibration, the influent
while Tony model,
Flameling and Petra
StefanVan Daal
Weijers
improved
were the model
involved structure and of
in the development performed
the modelthe model
and calibration,
provision while
of data. Tony
Ingmar Flameling
Nopens and Stefan
contributed Weijers
to the gap
filling. Jeroen Langeveld
were involved wrote the paper.
in the development of the model and provision of data. Ingmar Nopens contributed to the gap
filling. Jeroen
Conflicts Langeveld
of Interest: The wrote
authorsthedeclare
paper.no conflict of interest.

Conflicts of Interest: The authors declare no conflict of interest.

Water 2017, 9, 491 17 of 18

References
1. Benedetti, L.; De Baets, B.; Nopens, I.; Vanrolleghem, P.A. Multi-criteria analysis of wastewater treatment
plant design and control scenarios under uncertainty. Environ. Model. Softw. 2010, 25, 616–621. [CrossRef]
2. Rieger, L.; Gillot, S.; Langergraber, G.; Ohtsuki, T.; Shaw, A.; Takács, I.; Winkler, S. Guidelines for Using
Activated Sludge Models. In IWA Scientific and Technical Report No. 22; IWA Publishing: London, UK, 2012.
3. Rodriguez-Roda, I.; Sànchez-Marrè, M.; Comas, J.; Baeza, J.; Colprim, J.; Lafuente, J.; Cortes, U.; Poch, M.
A hybrid supervisory system to support WWTP operation: Implementation and validation. Water Sci. Technol.
2002, 45, 289–297. [PubMed]
4. Flores-Alsina, X.; Rodríguez-Roda, I.; Sin, G.; Gernaey, K. Multi-criteria evaluation of wastewater treatment
plant control strategies under uncertainty. Water Res. 2008, 42, 4485–4497. [CrossRef] [PubMed]
5. Nopens, I.; Benedetti, L.; Jeppsson, U.; Pons, M.N.; Alex, J.; Copp, J.B.; Gernaey, K.V.; Rosen, C.; Steyer, J.P.;
Vanrolleghem, P.A. Benchmark simulation model No 2: Finalisation of plant layout and default control
strategy. Water Sci. Technol. 2010, 62, 1967–1974. [CrossRef] [PubMed]
6. Maruejouls, T.; Lessard, P.; Wipliez, B.; Pelletier, G.; Vanrolleghem, P.A. A phenomenological retention tank
model using settling velocity distributions. Water Res. 2012, 46, 6857–6867. [CrossRef] [PubMed]
7. Martin, C.; Vanrolleghem, P.A. Analysing, completing, and generating influent data for WWTP modelling: A
critical review. Environ. Model. Softw. 2014, 60, 188–201. [CrossRef]
8. Gernaey, K.V.; Flores-Alsina, X.; Rosen, C.; Benedetti, L.; Jeppsson, U. Dynamic influent pollutant disturbance
scenario generation using a phenomenological modelling approach. Environ. Model. Softw. 2011, 26,
1255–1267. [CrossRef]
9. Flores-Alsina, X.; Saagi, R.; Lindblom, E.; Thirsing, C.; Thornberg, D.; Gernaey, K.V.; Jeppsson, U. Calibration
and validation of a phenomenological influent pollutant disturbance scenario generator using full-scale data.
Water Res. 2014, 51, 172–185. [CrossRef] [PubMed]
10. De Keyser, W.; Gevaert, V.; Verdonck, F.; De Baets, B.; Benedetti, L. An emission time series generator for
pollutant release modelling in urban areas. Environ. Model. Softw. 2010, 25, 554–561. [CrossRef]
11. Devisscher, M.; Ciacci, G.; Fé, L.; Benedetti, L.; Bixio, D.; Thoeye, C.; De Gueldre, G.; Marsili-Libelli, S.;
Vanrolleghem, P.A. Estimating costs and benefits of advanced control for wastewater treatment plants—The
MAgIC methodology. Water Sci. Technol. 2006, 53, 215–223. [CrossRef] [PubMed]
12. Talebizadeh, M.; Belia, E.; Vanrolleghem, P.A. Influent generator for probabilistic modeling of nutrient
removal wastewater treatment plants. Environ. Model. Softw. 2016, 77, 32–49. [CrossRef]
13. Bertrand-Krajewski, J.-L. Stormwater pollutant loads modelling: Epistemological aspects and case studies
on the influence of field data sets on calibration and verification. Water Sci. Technol. 2007, 55, 1–17. [CrossRef]
[PubMed]
14. Bertrand-Krajewski, J.-L.; Briat, P.; Scrivener, O. Sewer sediment production and transport modelling: A
literature review. J. Hydraul. Res. 1993, 31, 435–460. [CrossRef]
15. Bertrand-Krajewski, J.-L.; Chebbo, G.; Saget, A. Distribution of pollutant mass vs volume in stormwater
discharges and the first flush phenomenon. Water Res. 1998, 32, 2341–2356. [CrossRef]
16. Ashley, R.M.; Bertrand-Krajewski, J.L.; Hvitved-Jacobsen, T.; Verbanck, M. Solids in-Sewers; IWA Publishing:
London, UK, 2004.
17. Ashley, R.M.; Hvitved-Jacobsen, T.; Bertrand-Krajewski, J.-L. Quo vadis sewer process modelling? Water Sci.
Technol. 1999, 39, 9–22. [CrossRef]
18. Willems, P. Random number generator or sewer water quality model? Water Sci. Technol. 2006, 54, 387–394.
[CrossRef] [PubMed]
19. Willems, P. Parsimonious model for combined sewer overflow pollution. J. Environ. Eng. 2010, 136, 316–325.
[CrossRef]
20. Dembélé, A.; Bertrand-Krajewski, J.-L.; Becouze, C.; Barillon, B. A new empirical model for stormwater TSS
event mean concentrations (EMCs). Water Sci. Technol. 2011, 64, 1926–1934. [CrossRef] [PubMed]
21. Rousseau, D.; Verdonck, F.; Moerman, O.; Carrette, R.; Thoeye, C.; Meirlaen, J.; Vanrolleghem, P.A.
Development of a risk assessment based technique for design/retrofitting of WWTPs. Water Sci. Technol.
2001, 43, 287–294. [PubMed]
22. Gruber, G.; Winkler, S.; Pressl, A. Quantification of pollution loads from CSOs into surface water bodies by
means of online techniques. Water Sci. Technol. 2004, 50, 73–80. [PubMed]
Water 2017, 9, 491 18 of 18

23. Schilperoort, R.P.S. Monitoring as a Tool for the Assessment of Wastewater Quality Dynamics. Ph.D. Thesis,
Delft University of Technology, Delft, The Netherlands, 2011.
24. Schilperoort, R.P.S.; Dirksen, J.; Langeveld, J.G.; Clemens, F.H.L.R. Assessing characteristic time and space
scales of in-sewer processes by analysis of one year of continuous in-sewer monitoring data. Water Sci. Technol.
2012, 66, 1614–1620. [CrossRef] [PubMed]
25. 2000/60/EC. Directive of the European Parliament and of the Council Establishing a Framework for
Community Action in the Field of Water Policy (2000/60/EC). 23 October 2000. Available online: http:
//eur-lex.europa.eu (accessed on 3 July 2017).
26. Weijers, S.R.; de Jonge, J.; van Zanten, O.; Benedetti, L.; Langeveld, J.; Menkveld, H.W.; van
Nieuwenhuijzen, A.F. KALLISTO: Cost Effective and Integrated Optimization of the Urban Wastewater
System Eindhoven. Water Pract. Technol. 2012, 7. [CrossRef]
27. Benedetti, L.; Langeveld, J.; van Nieuwenhuijzen, A.F.; de Jonge, J.; de Klein, J.J.M.; de Flameling, T.;
Nopens, I.; van Zanten, O.; Weijers, S. Cost-effective solutions for water quality improvement in the
Dommel River supported by sewer-WWTP-river integrated modelling. Water Sci. Technol. 2013, 68, 965–973.
[CrossRef] [PubMed]
28. Langeveld, J.G.; Benedetti, L.; de Klein, J.J.M.; Nopens, I.; Amerlinck, Y.; van Nieuwenhuijzen, A.;
Flameling, T.; van Zanten, O.; Weijers, S. Impact-based integrated real-time control for improvement of the
Dommel River water quality. Urban Water J. 2013, 10, 312–329. [CrossRef]
29. Daal-Rombouts, P.; van Benedetti, L.; de Jonge, J.; Weijers, S.; Langeveld, J. Performance evaluation of a
smart buffer control at a wastewater treatment plant. Water Res. 2017. submit (under review).
30. Vrugt, J.A.; Ter Braak, C.J.F.; Clark, M.P.; Hyman, J.M.; Robinson, B.A. Treatment of input uncertainty
in hydrologic modeling: Doing hydrology backward with Markov chain Monte Carlo simulation.
Water Resour. Res. 2008, 44, 1–15. [CrossRef]
31. Vrugt, J.A.; Ter Braak, C.J.F.; Gupta, H.V.; Robinson, B.A. Equifinality of formal (DREAM) and informal
(GLUE) Bayesian approaches in hydrologic modeling? Stoch. Environ. Res. Risk Assess. 2009, 23, 1011–1026.
[CrossRef]
32. Keating, E.H.; Doherty, J.; Vrugt, J.A.; Kang, Q. Optimization and uncertainty assessment of strongly
nonlinear groundwater models with high parameter dimensionality. Water Resour. Res. 2010, 46, 1–8.
[CrossRef]
33. Leonhardt, G.; Sun, S.; Rauch, W.; Bertrand-Krajewski, J.-L. Comparison of two model based approaches for
areal rainfall estimation in urban hydrology. J. Hydrol. 2014, 511, 880–890. [CrossRef]
34. Van Daal-Rombouts, P.; Sun, S.; Langeveld, J.; Bertrand-Krajewski, J.-L.; Clemens, F. Design and performance
evaluation of a simplified dynamic model for combined sewer overflows in pumped sewer systems. J. Hydrol.
2016, 538, 609–624. [CrossRef]
35. Langeveld, J.G.; Clemens, F.H.L.R.; Van Der Graaf, J.H.J.M. Interactions within the wastewater system:
Requirements for sewer processes modelling. Water Sci. Technol. 2003, 47, 101–108. [PubMed]
36. Boogaard, F.; Lemmen, G. STOWA Stormwater Database: The Facts about the Quality of Stormwater Runoff ;
STOWA 2007-21; STOWA: Amersfoort, The Netherlands, 2007. (In Dutch)
37. Krebs, P.; Merkel, K.; Kühn, V. Dynamic Changes in Wastewater Composition during Rain Runoff.
In Proceedings of the 8ICUSD, Sydney, Australia, 30 August–3 September 1999; pp. 920–927.
38. Bruns, J.; von Regenwasserbehandlung, D.K.; bei Mischwasserzufluss, A. Stuttgarter Berichte zur
Siedlungswasserwirtschaft; Bd. 151. (Stuttgart, Univ. Diss., 1998); Oldenbourg: München, Germany, 1999.
39. De Mulder, C.; Flameling, T.; Langeveld, J.; Amerlinck, Y.; Weijers, S.; Nopens, I. Automating the Raw Data
to Model Input Process Using Flexible Open Source Tools. In Proceedings of the Frontiers International
Conference on Wastewater Treatment, Palermo, Italy, 21–24 May 2017.

© 2017 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).