Water Research 156 (2019) 264e276

Contents lists available at ScienceDirect

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

Evaluation of anaerobic digestion post-treatment options using an

integrated model-based approach
X. Flores-Alsina a, *, H. Feldman a, V.T. Monje a, P. Ramin a, K. Kjellberg b, U. Jeppsson c,
D.J. Batstone d, K.V. Gernaey a
a
Process and Systems Engineering Centr (PROSYS), Department of Chemical and Biochemical Engineering, Technical University of Denmark, Building, 229,
DK-2800 Kgs. Lyngby, Denmark
b
Novozymes A/S, Hallas Alle 1, DK-4400 Kalundborg, Denmark
c
Division of Industrial Electrical Engineering and Automation, Department of Biomedical Engineering, Lund University, Box 118, SE-221 00, Lund, Sweden
d
Advanced Water Management Centre (AWMC), The University of Queensland, St Lucia, Brisbane, Queensland, 4072, Australia

a r t i c l e i n f o a b s t r a c t

Article history: The objective of this paper is to present the main results of an engineering-research project dealing with
Received 3 December 2018 model-based evaluation of waste streams treatment from a biotech company. This has been extensively
Received in revised form done in domestic treatment systems, but is equally important, and with different challenges in industrial
16 February 2019
wastewater treatment. A new set of biological (activated sludge, anaerobic digestion), physicochemical
Accepted 18 February 2019
(aqueous phase, precipitation, mass transfer) process models and model interfaces are required to
Available online 4 March 2019
describe removal of organics in an upflow anaerobic sludge blanket (UASB) reactor plus either traditional
nitrification/denitrification (A1) or partial nitritation (PN)/anammox (ANX) (A2) processes. Model-based
Keywords:
Anammox
analysis shows that option A1 requires a decrease in digestion energy recovery (Erecovery) in order to have
Bacterial competition enough organic substrate for subsequent post NO3 reduction treatment (95 kWh.kg N
1). In contrast, A2
Biofilms in an aerobic granular sludge reactor allows for higher UASB conversion since N removal is carried out
Industrial wastewater autotrophically. The study also reveals that the addition of an aerated pre-treatment unit prior to the PN/
Physico chemical modelling ANX (A2) reactor promotes COD and H2S oxidation, CO2 and CH4 stripping, a pH increase (up to 8.5) and a
reduction of the risk of intra-granular precipitation as well as sulfide inhibition. Simulations indicate
clear differences regarding the microbial distribution/abundance within the biofilm in A2 when
comparing the two operational modes. Final results show the effects of different loading and operational
conditions; dissolved oxygen (DO), Total Suspended Solids (TSSop), energy recovery (Erecovery); on the
overall process performance; N removal, aeration energy (Eaeration), net energy production (Erecovery);
using response surfaces, highlighting the need of integrated approaches to avoid sub-optimal outcomes.
The study shows the benefits of virtual plant simulation and demonstrates the potential of model-based
evaluation when process engineers in industry have to decide between competing options.
© 2019 Elsevier Ltd. All rights reserved.

1. Introduction
Upflow Anaerobic Sludge Blanket (UASB) reactors provide a
sustainable and economic approach to treat high strength domestic
and industrial wastewaters (Batstone et al., 2015). These systems
can: 1) substantially reduce the organic load to be in accordance
with the effluent discharge limits, 2) produce biogas that can be
converted to energy (heat/electricity) to reduce the power
* Corresponding author.
E-mail address: xfa@kt.dtu.dk (X. Flores-Alsina).
demands within the treatment facility, and 3) handle a high volu-
metric load with a small spatial footprint (Van Lier, 2015). Never-
theless, the effluent of UASB digesters may contain high quantities
of nitrogen, sulfide and methane (Chen et al., 2016a, b). Nitrogen
plays an important role in the eutrophication of the urban catch-
ment and can be toxic for aquatic life (Tchobanoglous et al., 2003).
Sulfide causes pipe corrosion, health hazards and odour nuisance
(Zhang et al., 2008). Methane is a potent greenhouse gas with a
warming potential 34 times stronger than that of carbon dioxide
over a 100-year horizon (IPCC, 2013). The transformation/destruc-
tion/capture of nitrogen, sulfide and methane in different forms
and qualities should be achieved by adding downstream processes

https://doi.org/10.1016/j.watres.2019.02.035
0043-1354/© 2019 Elsevier Ltd. All rights reserved.
 X. Flores-Alsina et al. / Water Research 156 (2019) 264e276 265

Nomenclature SCa Calcium (ADM) (kmol.m
3) (ASM) (g.m
3)

SCl Chloride (ADM) (kmol.m
3) (ASM) (g.m
3)
AD Anaerobic digestion SCH4 Methane (ADM) (kg.m
3) (ASM) (g.m
3)
ADM Anaerobic Digestion Model SEtOH Ethanol (ADM) (kmol.m
3)
AER Aerobic zone (activated sludge reactor) Sfa Fatty acids (ADM) (kg.m
3)
Ai Alternative, post-digestion treatment option: i ¼ 1 SH2 Hydrogen (ADM) (kg.m
3)
traditional nitrification/denitrification, i ¼ 2 partial SHx S Inorganic total sulfides (ADM) (kg.m
3) (ASM)
nitritation/anammox (g.m
3)
ANOX Anoxic zone (activated sludge reactor) SK Potassium (ADM) (kmol.m
3) (ASM) (g.m
3)
ASM Activated Sludge Model SMg Magnesium (ADM) (kmol.m
3) (ASM) (g.m
3)
BOD Biochemical Oxygen Demand SNa Sodium (ADM) (kmol.m
3) (ASM) (g.m
3)
BSM Benchmark Simulation Model S N2 Nitrogen gas (ASM) (g.m
3)
CBIM Continuous Based Interfacing Methodology SNHx Inorganic ammonium nitrogen (ADM) (kmol.m
3)
COD Chemical Oxygen Demand (ASM) (g.m
3)
CSTR Continuous Stirred Tank Reactor SNO2 Nitrite (ASM) (g.m
3)
DO Dissolved Oxygen SNO3 Nitrate (ASM) (g.m
3)
Eaireation Energy consumption for aeration (kWh.d
1) S O2 Dissolved oxygen (ASM) (g.m
3)
EDSS Environmental Decision Support System Spro Total propionic acid (ADM) (kg.m
3)
Eheating Energy consumption for heating (kWh.d
1) SPOx Inorganic phosphorus (ADM) (kmol.m
3) (ASM)
Eheating,net Net energy consumption for heating (kWh.d
1) (g.m
3)
assuming that 1 kg CH4 produces 7 kWh of heat SSOx Sulfate (ADM) (kmol.m
3) (ASM) (g.m
3)
(resulting from the gas motor) Ssu Sugars (ADM) (kg.m
3)
Enet Net energy balancing (kWh.d
1) Sva Total valeric acid (ADM) (kg.m
3)
Epumping Energy consumption for pumping (kWh.d
1) SCi Scenario evaluated (i ¼ 1 to 3)
EQI Effluent Quality Index SEC Secondary clarifier
Erecovery Energy recovery (kWh.d
1) SIi Saturation Index, i ¼ CaCO3 and Ca3(PO4)2
Fe Iron SPLIT Splitter unit
G-L Gas liquid separator (UASB reactor) TKN Total Kjeldahl Nitrogen (g.m
3)
HRT Hydraulic retention time (h) TN Total Nitrogen (g.m
3)
I Ionic strength TSS Total Suspended Solids
IWA International Water Association TSSop Operational Total Suspended Solids (ASM) (g.m
3)
KH Henry's constant UASB Upflow Anaerobic Sludge Blanket
KLai Transfer coefficient, i ¼ oxygen, carbon dioxide, Vbiofilm Biofilm volume (m3) (for option A2)
nitrogen gas, ammonia and sulfide Vbulk Bulk volume (m3) (for option A2)
L Biofilm thickness VFA Volatile Fatty Acids
M Mixing section (UASB reactor) WRRF Water Resource Recovery Facility
MLE Modified Ludzack Ettinger process WWTS Wastewater Treatment System
N/DN Nitrification/Dentrification X Particulate compound
OCI Operational Cost Index Xac Acetate degraders (ADM) (kg.m
3)
ODE Ordinary Differential Equation XANX Anammox Bacteria (ASM) (g.m
3)
P Phosphorus XAOB Ammonium Oxidizing Bacteria (ASM) (g.m
3)
PCM Physico-Chemical Model Xbio Total biomass (ASM) (g.m
3) (ADM) (kg.m
3)
Pi Partial pressure of compound i XC4 Butyrate and valerate degraders (ADM) (kg.m
3)
PI Proportional Integral controller XCa3 ðPO3Þ 4 Calcium phosphate (ADM) (kmol.m
3) (ASM) (g.m
3)
PN/ANX Partial Nitritation/Anammox XCaðCO3 Þ Calcium carbonate (ADM) (kmol.m
3) (ASM) (g.m
3)
PreAD Pre-acidification tank before the AD reactor XDAMO1 Denitrifying Anaerobic Methane Oxidizing archaea
PWM Plant-Wide Model (ASM) (g.m
3)
Qair Airflow (activated sludge reactor, aerobic granular XDAMO2 Denitrifying Anaerobic Methane Oxidation bacteria
reactor) (ASM) (g.m
3)
Qintr Internal recycle flow between AER3 to ANOX1 (for XEtOH Ethanol Oxidizing Bacteria (ADM) (kg.m
3)
option A1) (m3.d
1) XI Inert particulate organics (ADM) (kg.m
3) ASM
Qr External recirculation flow between SEC to ANOX1 (g.m
3)
(for option A1) (m3.d
1) Xli Lipids (ADM) (kg.m
3)
Qw Waste flow from SEC (for option A1) (m3.d
1) XMOB Methane Oxidizing Bacteria (ASM) (g.m
3)
R1 Expanded sludge bed (AD reactor) XNOB Nitrite Oxidizing Bacteria (ASM) (g.m
3)
R2 Polishing section (AD reactor) XOHO Ordinary Heterotrophic Bacteria (ASM) (g.m
3)
REAC Aerobic granular sludge reactor (physical unit) where Xorg Particulate organics (ASM) (g.m
3)
PN/ANX takes place Xpr Proteins (ADM) (kg.m
3)
S Soluble compound Xpro Propionate degraders (ADM) (kg.m
3)
Saa Amino acids (ADM) (kg.m
3) XS Biodegradable particulate organics ASM (g.m
3)
Sac Total acetic acid (ADM) (kg.m
3) XSRB Sulfate reducing bacteria (ADM) (kg.m
3)
SiC Inorganic carbon (ADM) (kmol.m
3) (ASM) (g.m
3) XS0 Mineral sulphur (ADM) (kg.m
3) (ASM) (g.m
3)
Sbu Total butyric acid (ADM) (kg.m
3)
266 X. Flores-Alsina et al. / Water Research 156 (2019) 264e276
zmax Maximum biofilm thickness (in UASB and aerobic
granular reactors)
z Radial distance within the biofilm (m)
Z Inorganic species (calculated with the PCM)
when designing Wastewater Treatment Systems (WWTS) (Vela
et al., 2015).
Mathematical models are currently applied in normal water
engineering practice to evaluate competing design treatment op-
tions (Flores-Alsina et al., 2012; Benedetti et al., 2010). In addition,
models can also be used for benchmarking (Copp, 2002; Jeppsson
et al., 2007), diagnosing (Rodriguez-Roda et al., 2002), and opti-
mizing (Rivas et al., 2008; Feldman et al., 2018) water systems, as
well as for teaching purposes (Hug et al., 2009). Since the publi-
cation of the International Water Association (IWA) Activated
Sludge Models (ASM1, 2d & 3) (Henze et al., 2000) and the
Anaerobic Digestion Model No. 1 (ADM1) (Batstone et al., 2002), the
field has been rapidly expanding. Special efforts have been devoted
to Plant-Wide Modelling (PWM) in recent years. The main idea in
PWM is that the WWTS should be considered as an integrated
process, where the different units comprising the flow diagram are
linked together and need to be operated and controlled not as in-
dividual unit operations, but taking into account all the interactions
amongst the processes (Skogestad, 2000; Jeppsson et al., 2013).
One of the most difficult aspects in plant-wide modelling is
prediction and interaction of aquatic chemistry states, which are
not normally considered in an efficient way in biological models.
Recent advances by the IWA Task Group on Physico-Chemical
Modelling (PCM) (Batstone et al., 2012) have boosted this field of
research, particularly because considering PCM in addition to bio-
logical process models allows predicting the close interactions
between the phosphorus (P), sulphur (S) and iron (Fe) cycles (Solon
et al., 2017). Special aspects, such as model interfacing, have to be
taken into account (Batstone et al., 2015), unless integrated
modelling methodologies are used (Barker and Dold, 1997; Grau
et al., 2007; Ekama, 2009; Barat et al., 2013; Ferna ndez-Are valo
et al., 2017). PWM requires elemental mass balance verification
and continuity checking for all the components comprising the
model (Volcke et al., 2006a; Zaher et al., 2007; Nopens et al., 2009).
Therefore, COD, C, N, P, Fe and S need to be balanced across the
interface (Flores-Alsina et al., 2016). The main advantage of using
an interface-based approach with respect to other integrated
methodologies is that the original model structure can be used,
thus there is no need to consider all the state variables for each
simulated unit process, with a reduction of the required compu-
tational power and better overall model transparency (Grau et al.,
2009).
In this paper, we present a model-based feasibility study of
different anaerobic digestion post-treatment options in a biotech-
nology company using a PWM. The main motivation of the study is
to analyse how the selected technologies may affect the fate of
nitrogen, methane and sulfide compounds leaving an industrial AD
system. Special aspects such as energy expenditures and potential
(energy) recovery through the entire flow diagram will be
accounted for. This will be done using a previously published multi-
scale mathematical model approach describing the competition
between sulfate reducing bacteria and methanogens, the addition
of ethanol as a separate model component, the effect of ionic
strength/activity corrections and the formation of multiple mineral
precipitates (Feldman et al., 2017). The above is linked to two
different reactor designs (activated sludge/granular) representing
two post treatment options where a new model is developed. This
model describes nitrification, denitrification (autotrophic/
ZCO2 Carbon dioxide (PCM) (kmol.m
3)
ZHx S Free hydrogen sulfide (PCM) (kmol.m
3)
ZHNO2 Free nitrous acid (PCM) (kmol.m
3)
ZNH3 Free ammonia (PCM) (kmol.m
3)
heterotrophic) and the main (bio)chemical processes involving
methane and sulphur transformations. The entire system
(AD þ post treatment) is simulated simultaneously. The main
novelties of this contribution can be summarized as follows:
1) A plant-wide model incorporating a UASB reactor þ two post-
treatment alternatives within an industrial context. Published
studies considering plant-wide aspects have focused on virtual
case studies (Volcke et al., 2006b; Flores-Alsina et al., 2014).
When full-scale facilities were modelled, the type of wastewater
was urban (Kazadi-Mbama et al., 2016). Indeed, industrial
wastewater has some drivers and challenges, including elevated
sulfide, high strength inputs and very different system config-
urations, which make state-of-the-art-models for municipal
wastewater systems less suitable, and often additional exten-
sions are needed (Feldman et al., 2017).
2) Simultaneous simulation of two types of biofilm systems (anaer-
obic/aerobic). Previous investigations have often been focused
on studying process units (AD, post treatment) separately
(Batstone et al., 2015; Chen et al., 2016a, b; Winkler et al., 2014;
Castro-Barros et al., 2018) but never combined.
3) A biofilm model implementation allowing the consideration of
control aspects. This has been a particular drawback when ana-
lysing aerobic/anaerobic granular systems in the past (Volcke
et al., 2010, 2012; Ni et al., 2009), where the state-of-the-art
software used for these studies (Reichert, 2001) did not allow
for the inclusion of sophisticated process control elements
(Rosen et al., 2008).
4) Integration of physico-chemical aspects within (aerobic) biofilms
systems. Previous works have been focused on granular micro-
bial composition (Pe rez et al., 2014; Poot et al., 2016), but there is
no study where rigorous aqueous phase chemistry has been
included when modelling aerobic biofilms (Vangsgaard et al.,
2013).
5) A holistic evaluation approach using net energy balances (UASB
recovery e post treatment energy demand) that are applied under
different operational conditions. This is closely linked to the first
point, where there is a lack of full-scale studies using integrated
models and global criteria (Solon et al., 2017; Ferna ndez-Are valo
et al., 2017).
The paper details the development of the new PWM developed
for this study by presenting sequentially the elements of which it is
comprised as well as highlighting the integration/interfacing as-
pects. The capabilities/potentials of the proposed approach are
illustrated by means of several simulated scenario analyses. Lastly,
opportunities and limitations that arise from utilization of the new
models are discussed in detail.
2. Methods
Sections 2.1.1 and 2.1.2 describe the processes and state variables
included in both the Anaerobic Digestion (ADM) and the Activated
Sludge (ASM) models, respectively. Additional aspects, with special
emphasis on linking the ADM and ASM with a Physico-Chemical
Model (PCM) (Section 2.2), are described in Section 2.3 (Model
integration). Implementation details for different types of reactor
designs: UASB, activated sludge and aerobic granular, are detailed
 X. Flores-Alsina et al. / Water Research 156 (2019) 264e276
in Sections 2.4.2, 2.4.3 and 2.4.4.
2.1. Biological models
2.1.1. Anaerobic digestion model (ADM)
The ADM1 version, implemented in the plant-wide context
provided by the Benchmark Simulation Model No. 2 (BSM2)
(Batstone et al., 2002; Rosen et al., 2006) is extended with P, S and
Fe interactions (Flores-Alsina et al., 2016). It also includes ethanol
(SEtOH ) degradation, which is modelled assuming a specific group of
microorganisms (XEtOH ), which ends up producing hydrogen (SH2 )
and acetate (Sac ) (Batstone et al., 2004). Mass balances, stoichi-
ometry coefficients and continuity verification of the model can be
found Supplemental Information section (xls files).
2.1.2. Activated sludge model (ASM)
Reaction rates describing organic carbon (C) and nitrogen (N)
removal are described as in Vangsgaard et al. (2012). The model
accounts for Ammonium/Nitrite Oxidizing Bacteria (XAOB /XNOB ),
Anammox Bacteria (XANX ) and Ordinary Heterotrophic Bacteria
(XOHO ). The default implementation is upgraded to include
methane (SCH4 ) and sulphur species (SHx S / XS0 / SSOx ) conversion
processes. Hence, Methane Oxidation Bacteria (XMOB ), Denitrifying
Anaerobic Methane Oxidation (DAMO) archaea (XDAMO1 ) and bac-
teria (XDAMO2 ) are the main responsible organisms for biological
SCH4 removal (Winkler et al., 2014; Daelman et al., 2014). Biological
conversion of SHx S / XS0 / SSOx is carried out by Sulphur Oxidation
Bacteria ðXSOB Þ involving different electron acceptors (SO2 / SNO3 /
SNO2 ). Further information can be found in (Chen et al., 2016a, b).
The compounds SHx S and XS0 can also be oxidized chemically using
a two-step reaction (from SHx S to XS0 and finally to SSOx ) as
described in Gutierrez et al. (2010). Mass balances, stoichiometry
coefficients and continuity verification of this part of the model can
also be found in the Supplemental Information section.
2.2. Physico-chemical models (PCM)
2.2.1. Aqueous phase chemistry þ precipitation model
A common physico-chemical framework is implemented to both
ADM and ASM. The model simulates the acid-base system and is
thereby able to predict pH (Solon et al., 2015). It also corrects for
ionic strength (I) via the Davies approach to consider chemical
activities instead of molar concentrations and hence perform all the
calculations under non-ideal conditions (Flores-Alsina et al., 2015).
Multiple mineral precipitation potential is based on Saturation In-
dex calculations (SI) as described in Kazadi Mbamba et al. (2015a,
b).
2.2.2. Gas-liquid transfer
Mass transfer between the liquid and the gas phase is accounted
for selected compounds (i ¼ SO2 , SN2 , ZCO2 , ZH2 S , ZNH3 , SCH4 and SH2 ).
The transport rates are formulated as a function of the difference
between the saturation concentration and the actual concentration
of the gas dissolved in the liquid (Batstone et al., 2012). The satu-
ration concentration of the gas in the liquid is given by Henry's law
of dissolution, which states that the saturation concentration is
equal to the product of Henry's constant (KH) multiplied by the
partial pressure of the gas (Pi). The latter will be different in closed
(UASB) and open (N/DN and PN/ANX) reactors (Rosen et al., 2006).
The mass transfer rate constant (KLai) is calculated for each gaseous
component as described in Musvoto et al. (2000).
267
2.3. Model interfacing
2.3.1. ADM to ASM interface
In this study the continuous-based interface methodology CBIM
is used to ensure continuity in the mass balances and avoid
component leaks (Volcke et al., 2006a; Nopens et al., 2009). All the
ADM organic states, such as carbohydrates, proteins and lipids (Xpr ,
Xli , Xch ), sugars, amino acids and fatty acids (Saa , Sfa , Ssu ) and volatile
fatty acids (Sac , Spro , Sbu , Sva ) are converted into ASM general/
lumped biodegradable organic soluble (SS ) and particulate (XS )
materials. In addition, immediate decay of the AD biomass (Xbio Þ to
XS and organic inert compounds (XI Þ is assumed. There is no vari-
ation of cations/anions before and after the interface. A compre-
hensive description with detailed explanation of the involved
processes, conversions and mass balance verification can be found
in Flores-Alsina et al. (2016).
2.3.2. ASM/ADM to PCM interface
The outputs of the ASM/ADM interface at each integration step
are used as inputs for the PCM module to estimate pH and ion
speciation/pairing (PMC module works as a sub-routine). The
precipitation/stripping/inhibition equations are embedded within
the AD/AS ordinary differential equation (ODE) model structure
and are therefore included in the overall mass balance. This is
especially complicated in granular systems (UASB þ PN/ANX op-
tion), where this communication is established for each discretized
layer within the biofilm (see Sections 2.4.2 and 2.4.4). For example,
the PCM predicts the quantity of free ammonia (ZNH3 ) and free
nitrous acid (ZHNO2 ) used by XAOB and XNOB as substrates for growth
in the ASM implementation. In a similar way, calculated hydrogen
sulfide (ZH2 S ) and ammonia (ZNH3 ) concentrations are used for
calculating both stripping rates and inhibition functions in both the
ADM and ASM approaches, according to Batstone et al. (2002) and
Kalyuzhnyi and Fedorovich (1998). Finally, the calculation of the SI
is the driving force to trigger the precipitation reactions, which in
turn are based on activities (Kazadi-Mbamba et al., 2015a, b).
2.4. Additional elements
2.4.1. Data measuring campaign and influent generation
The influent flow rate/mass loadings into the industrial AD are
144 m3 h
1, 1600 kg COD.h
1, 80 kg N.h
1 and 9 kg P.h
1. The
influent S:COD ratio is 0.025 kg S.(kg COD)
1. It is important to
highlight that the influent to be treated is industrial wastewater,
not waste activated sludge. Samples were taken at the outlet of the
pre-acidification tank (PreAD), a unit that is located before the
studied anaerobic granular bioreactor to promote the formation of
VFA. Data on influent/effluent cations (Naþ, Kþ, Mg2þ, Ca2þ) and
anions (Cl
, SO2

4 ) concentrations are available. Analyses were done
using Standard Methods (APHA, 2012), using 1-day composite
samples. Influent fractionation and generation of additional
influent dynamics were carried out as stated in Feldman et al.
(2017). Interested readers are referred to the Supplemental infor-
mation section where specific information about the influent
characteristics can be found.
2.4.2. USAB reactor model
The plant under study is located in Kalundborg in the North
Western part of Zealand (Denmark) and treats the wastewater (not
sludge) from Novo Nordisk and Novozymes, two of the biggest
biotechnology industries in Denmark. Reactor design is based on
the BIOPAQ®IC technology (Paques, the Netherlands), an advanced
high-rate granular anaerobic digester. The reactor is comprised of
four parts: 1) a mixing section (M); 2) an expanded sludge bed (R1);
3) a polishing section (R2); and, 4) a gas-liquid separator (G-L) (see
268 X. Flores-Alsina et al. / Water Research 156 (2019) 264e276
Fig. 1. Schematic representation of the system under investigation: UASB reactor þ two digestion post treatment options. A1 ¼ conventional nitrification/denitrification system,
A2 ¼ PN/ANX system (with pre-treatment).
Fig. 1 for details). The multi-scale approach described in Feldman
et al. (2017) is used to describe reactor hydrodynamics using mul-
tiple Continuous Stirred Tank Reactors (CSTRs), a granule size dis-
tribution and biofilm growth. More specifically, the biofilm model is
based on the one-dimensional approach reported in Wanner et al.
(2006). The model contains both soluble (S) and particulate (X)
state variables. The mass balance assumes that the transport of
soluble compounds is governed solely by (homogenous) diffusion,
whereas movement of particulate compounds takes place by con-
vection (Saravanan and Sreekrishnan, 2006). Biofilm thickness (L) is
given as the radial distance (z) from the centre to the surface of the
granule and varies due to two phenomena: 1) the net growth of the
particulate species and 2) detachment from the biofilm surface
(Lackner et al., 2008). The resulting system of partial differential
equations (PDEs) is solved using the method of lines (Press et al.,
2007).
2.4.3. Conventional N/DN reactor model
The first configuration (A1) is based on a modified Ludzack-
Ettinger (MLE) design. The plant is described as 5 CSTRs in series
and one secondary sedimentation tank (SEC) (see Fig. 1 for details).
The behaviour of the SEC is approximated by using the double-
exponential settling function proposed in Tak acs et al. (1991).
Tanks 1 and 2 are anoxic (ANOX1, 2). Tanks 3, 4 and 5 (AER1, 2 and
3) are aerobic. All reactors are assumed to have the same volume.
AER3 and ANOX1 are linked by means of an internal recycle (QINTR).
Literature values were used to determine the hydraulic retention
time (HRT) and TSSop (Tchobanoglous et al., 2003). A PI controller
regulating the airflow (Qair) to maintain a constant dissolved oxy-
gen (DO) concentration (set-point ¼ 2 g (-COD).m
3) is imple-
mented in AER2. Proportional controllers are used to determine Qair
in AER1 and AER3 (see Gernaey et al., 2014) according to AER2. The
DO sensor response time (1 min), sensor delay and white noise are
included in the sensor/actuator models in order to avoid creating a
simulation of unrealistic ideal control applications. Finally, the
aeration system (Qair), is defined with significant dynamics,
assuming a response time of 4 min for the blowers. More info about
sensor and actuator models can be found elsewhere (Rieger et al.,
2003; Rosen et al., 2008).
2.4.4. Aerobic granular sludge reactor model
A similar approach as described in Section 2.4.2 is used to
describe the aerobic granular sludge reactor (A2). For simplicity
purposes, the REAC is modelled by assuming one single CSTR and
one single granule size (zmax) type (additional information in
Vansgaard et al. (2012)). The exact same strategy as described
above predicts competition for substrate/space within the biofilm
by the different microorganisms accounted for by the ASM. An ideal
separator (SEP) avoids that the granules leave REAC. An aerated
tank (PRE) and a splitting unit (SPLIT) are added before REAC, and
act as a pre-treatment using the ASM as in the conventional N/DN
reactor (Jeppsson et al., 2007) (see Fig. 1 for details). The objective of
PRE is to adjust COD/N ratios þ pH as well get rid of potentially
troublesome compounds (H2S) that can hinder Anammox biomass
activity. Literature values were used to determine the hydraulic
retention time (HRT) and TSSop (Lackner et al., 2014). A default
strategy based on a simple PI DO loop with similar sensor/actuator
characteristics (noise, delay, response time) as described in Section
2.4.3 is implemented (set-point ¼ 0.25 g (-COD).m
3). Note the
difference between REAC (physical unit where PN/ANX takes place)
and A2 (alternative comprising both PRE þ REAC).
2.4.5. Evaluation criteria
To assess the performance of the different treatment options, a
set of evaluation criteria is necessary (Jeppsson et al., 2013). Different
effluent quality criteria, such as Total Kjeldahl Nitrogen (TKN) and
Total Nitrogen (TN), are calculated from ASM state variables (Henze
et al., 2000) to assess the pollution removal efficiency of the system.
Energy recovery (Erecovery) in the UASB is calculated by using the
energy content of the methane gas (50.014 MJ (kg CH4)
1) and
assuming a 43% efficiency for electricity generation. Aeration energy
(Eaeration) is calculated imposing an energy expenditure equal to
 X. Flores-Alsina et al. / Water Research 156 (2019) 264e276 269

1.8 kg (-COD) (kWh)
1. Pumping energy (Epumping) is calculated as 3. Results

the weighted averaged sum of internal (QINTR) and external recycle
(QR) and waste flow (QW). Epumping also quantifies the quantity of
particulate material removed in PRE (for alternative A2) via under-
flow. The heating energy (Eheating) is calculated from the quantity of
water entering into the UASB. The net energy heating demand
(Eheating,net) is met by the heat generated from the assumed gas used
for electricity production from biogas (1 kg CH4 produces 7 kWh of
heat). The difference between Erecovery - Eaeration e Epumping e Ehea-
ting,net is the net energy generated (Enet). Further information can be
found in Gernaey et al. (2014).
2.5. Simulation scenarios
Three different scenarios are considered in this work (see
Table 1). The conditions for the first scenario (SC1) are selected to
study two potential options to treat the pollution loads from the
UASB effluent. The first technology (A1) is a conventional activated
sludge configuration where traditional nitrification (N)/denitrifi-
cation (DN) processes take place. The alternative configuration (A2)
includes a granular aerobic reactor based on the partial nitritation
(PN)/Anammox (ANX) process. In this particular scenario, the
compromise between energy recovery in the UASB and denitrifi-
cation potential is also studied in both post treatment cases by
conducting integrated simulations, i.e. UASB model and post-
treatment configuration model are solved simultaneously. The
second scenario (SC2) explores the effect of PRE on A2. Special
emphasis is placed on evaluation of influent characteristics, process
performance, biofilm microbial distribution and the risk of intra-
granule precipitation. The third scenario (SC3) explores variations
on the overall process performance (N removal, Eairation, Enet) when
selected operational/design/loading conditions for the UASB þ A2
treatment train (PN/ANX þ pre-treatment) are modified (DO, VSSOP,
Erecovery). All dynamic simulations start from steady state condi-
tions (1000 days) to eliminate bias due to the selection of initial
conditions (Gernaey et al., 2014). The sensor and actuator models
are only used during dynamic simulations.
Table 1
Definition of the different scenarios and assumed design/operational conditions.
3.1. Post-anaerobic digestion treated water options evaluation
3.1.1. Default operational conditions
Table 2 shows steady state UASB influent data þ model pre-
dictions for the proposed post-digestion treatment options (A1, A2).
The multi-scale approach, parameter values and full-scale data
reported in Feldman et al. (2017) are used to obtain the anaerobic
digester results represented in columns I and II (see Sections 2.4.1
and 2.4.2.). Model interfaces and the ASM implementation in the
aerobic granular reactor yield columns III and IV (see Sections 2.4.3
and 2.4.4). More information about model interfaces and how COD,
C, N and P are maintained when leaving the UASB reactor and
entering the subsequent post treatment options (for subsequent
post-treatment evaluation) are provided in the Supplemental in-
formation section. Simulation results displayed in Table 2 indicate
substantial differences regarding N removal efficiencies between A1
and A2 (74 and 90% respectively) while COD, CH4 and S removal are
almost the same. These differences are attributed to the fact that
scenario A1 fails to complete SNO3 / SNO2 denitrification due to
insufficient carbon source (SS ). Option A2 can reduce SNO2 to SN2
since this process is done autotrophically by XANX (and Salk is not a
limiting factor). The surplus heat that may be produced during
electricity generation is for heating the anaerobic digester, there-
fore Eheating,net is zero. Regarding Erecovery Table 2 shows that at
default operational conditions of the UASB reactor it is possible to
recover 149.2 MWh.d
1. This corresponds to 10,800 m3d-1 of CH4
(78% of the total produced biogas). When it comes to Eaeration, A1
would require 5652 kWh.d
1 or 3.95 kWh/kg Nremoved. These esti-
mations are close to the values reported in Lackner et al. (2014).
Additionally, around 102 kWh.d
1 is necessary to reach the
required pumping rates of typical MLE configurations (QINTR, QR
and QW). This leads to an Enet of 143.5 MWh.d
1. On the other hand
in A2, partial nitrification occurs and the required airflow is much
lower (2700 kWh.d
1 or 1.75 kWh/kg Nremoved) compared to A1
(even though A2 includes AE in both PRE and REAC). These values
correspond again to literature references where reported aeration
energy values for PN/ANX range from 1.05 to 4.7 kWh/kg Nremoved
(Lackner et al., 2014). This makes the operation of A1 significantly

UASB A1 (conventional N/ A2 (PN/ANX granular Table 2

DN) reactor) Performance evaluation of the different treatment trains (steady state simulation).
Sc1: Evaluation of post digestion options
Influent UASB A1 (conventional A2 (PN/ANX
HRT (h) 6e18 24þ8b 8þ12c
N/DN) granular
TSSop 50/35/15a 4 24d
reactor)
(kg.m
3)
zmax (m) 0.004/0.002/ e 0.0005 Effluent quality
0.0005a Kg COD.h
1 1600 450 117 120
DO (g.m
3) 0 2 0.25 Kg N.h
1 80 80 21 8
Sc2: Exploration the effect of pre-treatment (PRE) on A2 Kg CH4.h
1 6a <0.1 <0.1
HRT (h) 12 (8) þ 12c Kg S.h
1 43 19b <0.1 <0.1
TSSop 50/35/15a 24d Energy recovery, expenditures and net balancing
(kg.m
3) Erecovery (MWh.d )
1
149.2
zmax (m) 0.004/0.002/ 0.0005 Eheating/Eheating,net (MWh.d
1) 73c/0 d – e
0.0005a Eaireation (kWh.d
1) 5652 2700
DO (g.m
3) 0 0.25 Epumping (kWh.d
1) 102 20
Sc3: Local/Global evaluation of the overall process performance on A2 Enet (MWh.d
1) 143.5 146.5
HRT (h) 6e18 8þ12c a
Kg CH4.h
1 only represents the dissolved fraction in the liquid phase (UASB
TSSop 50/35/15a 15e30
effluent).
(kg.m
3) b
Kg S.h
1 only represents the dissolved fraction of reduced S forms (H2S/HS
) in
zmax (m) 0.004/0.002/ 0.0005
the liquid phase (UASB effluent).
0.0005a c
Eheating is energy input that is needed to heat the inlet flow to the UASB until it
DO (g.m
3) 0 0.1-.1.5
reaches the desired operational temperature (T ¼ 35 C).
a d
The reactor is discretized vertically in three sections: M, R1 and R2. Eheating,net is quantified assuming that 1 kg CH4 produces 7 kWh of heat
b
Includes the HRT of the bioreactor and the secondary settler. (resulting from the gas motor) from the quantity of methane produced Eheating;net ¼
c
Includes the HRT of the PRE unit. maxð0; Eheating
7MPÞ where MP is methane (gas) production (kgCH4.d
1).
d
This is translated to 20% of granule occupation within REAC.
270 X. Flores-Alsina et al. / Water Research 156 (2019) 264e276

more expensive compared to A2 (Enet is up to 146.5 MWh.d
1).
Dynamic simulations show again the differences in N removal
performance and how option A1 is not capable to remove SNO3 /
SNO2 due to insufficient available carbon source (see Fig. 2, row II).
There are substantial differences in SO2 levels for both options. In
option A1, all SNHx has to be transformed to SNO3 . Therefore, the SO2
demand is very high. In option A2, only a part of the incoming SNHx
is oxidized to SNO2 thereby reducing the aeration requirements
substantially (see Eaeration in Table 2). It is important to notice the
effect of the defined noise/delay level in the sensor and actuator
model. These model additions, e.g. the sensor models, provide a
more realistic PI response and hence contribute to avoiding having
unrealistic (¼ideal) control actions in the simulation (see Fig. 2, row
I). Both A1 and A2 options achieve almost complete removal of SCH4
and SHx S at the end of the treatment (>99% removal) (see rows 4
and 5). Stripping is the main mechanism for SCH4 removal in the two
evaluated options. Stripping in A1 takes place in the AER section,
while in A2 it is in the PRE section. The predicted activity of XMOB is
zero for A1 and A2. Nevertheless, there is a minor contribution of
XDAMO1 / XDAMO2 in SCH4 conversion for A2. Similar results were
reported in Daelman et al. (2014) and Winkler et al. (2014).
Regarding SHx S , this compound is (bio)chemically re-oxidated to
XS0 / SSOx (see row 6) in both A1 and A2. The relatively high oper-
ational pH (see Table 1) ensures a relatively low ZH2 S stripping, i.e.
the equilibrium is moved towards ZHS
(Stumm and Morgan, 1996).
3.1.2. Modification of the UASB reactor conversion efficiency
In this section, the balance between organics conversion within
the UASB reactor versus energy consumption and N removal in both
aerobic reactors (A1 & A2) is studied in more detail. In order to do
this, an increase/decrease of conversion efficiency is imposed by
changing the UASB reactor volume. A pH control ensured the same
acid-base conditions. This modified conversion efficiency, com-
bined with alternatives A1 and A2 respectively, is simulated using
the PWM presented in this paper (see Table 1). Fig. 3a and b shows
the changes in the quantity of SS leaving the reactor (from 240 to
100 kg h
1) and the quantity of methane produced (from 9500 to
11,400 m3 d
1). The latter is used to calculate the potential Erecovery

Fig. 2. UASB influent (column I), effluent (column II), data/model predictions and simulation results of the proposed post-digestion treatment options (columns III & IV).
A1 ¼ conventional nitrification/denitrification system, A2 ¼ PN/ANX system (including pre-treatment).
 X. Flores-Alsina et al. / Water Research 156 (2019) 264e276 271

Fig. 3. UASB capacity analysis (a,b) and effect on the denitrification efficiency of the two evaluated options (c,d). A1 ¼ conventional nitrification/denitrification system, A2 ¼ PN/ANX
system (with pre-treatment). On the two top plots, also experimental data (plant measurements) are included (marked with x). The bar centred on the figure, represents the default
operational conditions (see Table 2). Bars to the right and to the left of the default condition bars represent an increase and decrease of the AD conversion efficiency, respectively.

(see section 2.4.5). Fig. 3c and d shows the nitrogen removal effi-
ciency for the two evaluated options considering the different UASB
reactor capacities. It is important to highlight that in all cases, the
total quantity of nitrogen to treat is the same (80 kg N.h
1).
When the UASB reactor volume is lowered, higher N removal
rates were achieved in the option A1 (up to þ 67% of the total
incoming N). This is attributed to the higher quantity of organics
arriving from the AD to the MLE plant, and therefore more carbon
source is available for the traditional denitrification process. As a
direct result, there is a decrease of 11% in the Erecovery. When the
UASB reactor conversion is higher (up to þ 5% in Erecovery) by
increasing its volume it has negative effects (¼ not sufficient electron
donor) on nitrogen removal due to denitrification failure (down to
34% N removal). Using the numbers generated in this case study, we
can conclude that each additional kg of nitrogen removed in the post
treatment will require a decrease in the digester performance
equivalent to 95 kWh.kg N
1. More details are provided in Fig. 3c.
Regarding option A2, simulation results show that a higher
destruction of organics in the UASB reactor brings many benefits to
the overall plant performance. Firstly, the quantity of energy re-
covery is higher (up to þ 5% in Erecovery). Secondly, Eaeration in REAC
is slightly lower due to a minor decrease of the oxygen demand
required to remove the arriving COD (Eaeration changes from 1200 to
1100 kWh.d
1) (Cao et al., 2017). As a result Enet is more favourable
(from 146.5 to 153.9 MWh.d
1). Thirdly, the quantity of nitrogen
removed is slightly increased (from 80 to 84%) due to a lower
competition with XOHO giving advantage to XANX (Ni et al., 2009).
More details are provided in Fig. 3d.
3.2. Effect of pre-treatment on PN/ANX performance
3.2.1. PN/ANX influent characteristics
This section describes the effect on the overall process perfor-
mance when the aerated pre-treatment (PRE) unit is utilised (or
not) before the PN/ANX reactor (options A2,1 and A2,2) (see Section
2.4). The PRE consists of a CSTR with a residence time of 8 h, which
is aerated (see Table 1). Fig. 4 shows the UASB influent (see Section
2.4.1), the effluent (see Section 2.4.2.) and the PRE (see Section
2.2.4) model predictions as well as the measured data (when these
were available). Computer simulations show that adding the PRE
promoted a COD decrease (organic matter and H2S oxidation) in the
UASB effluent, CH4/CO2 stripping and a pH increase (up to pH 8.5).
High pH also promotes the formation of inorganic precipitates
(XCaCO3 and XCa3 ðPO4 Þ2 ), which decreases the quantity of soluble Ca
and P in the effluent. Since compounds containing Mg are present
in under-saturated conditions, no precipitates of Mg compounds
are formed (reflected as struvite in Fig. 4.
3.2.2. P/ANX effluent characteristics and granule structure/
properties
Fig. 5 reveals that the most immediate effect of PRE is a sub-
stantial reduction of the H2S inhibition (and consequently SNHx
conversion). Both A2,1 and A2,2 show that the centre of the granule
is inactive due to the high concentration of inert material (XI )
resulting from biomass death. The biomass (Xbio ) and organics
(Xorg ) concentrations increase for an increasing radial distance (z),
i.e. the closer to the surface, the higher the biomass concentration.
272 X. Flores-Alsina et al. / Water Research 156 (2019) 264e276

Fig. 4. UASB influent (row I), data (circles)/model predictions and simulation (solid lines) results obtained when including a PRE unit (rows II and III). A2,1 ¼ PN/ANX þ pre-
treatment, A2,2 ¼ PN/ANX without pre-treatment. Since the scenario A2,2 does not include pre-treatment, the influent of the PN/ANX is in that case exactly the same as the effluent
of the UASB reactor (where measurements were available).

More specifically, a high portion of XANX is placed in the inner zone
and this fraction decreases as the radius of the granule increases.
The presence of XAOB and XOHO is comparatively higher in the outer
layers of the biofilm. This distribution corresponds to the degree of
SO2 penetration within the biofilm.
The main differences between A2,1 and A2,2 rely on the relative
presence of XOHO , XDAMO and XSOB . This is attributed to the fact that
part of the COD, H2S and CH4 is removed in the PRE. Finally, PRE has
an effect on the weak acid-base chemistry of both the bulk and the
granule. As a consequence of including the PRE, the risk of having
intra-granular precipitation for calcite (SIcalc) and calcium phos-
phate (SICaP) is reduced. In both cases, the system is under-
saturated for struvite. Surprisingly, no pH gradient between the
biofilm and the bulk phase is predicted while some authors
observed otherwise (Lemaire et al., 2009). Other simulation studies
have shown that pH gradients are increased as function of the
granule size (Feldman et al., 2017). In this particular case, the flat pH
profiles can be justified due to the relatively low SNO2 concentra-
tions and the high buffer capacity of the system.
3.3. Variation in loading/operational conditions
3.3.1. DO set-point, TSSop and Erecovery
Response surfaces depicted in Fig. 6 show the effect of changing
the operational conditions on: 1) N removal (effluent SNHx /N
denitrified), 2) aeration energy (Eaeration), and 3) net energy balance
(Enet) when both UASB reactor þ A2 (PN/ANX þ pre-treatment) are
simulated simultaneously. Computer results show that optimal DO
levels for N removal ranges from 0.25 > DO < 0.5 g m
3. At lower
values (DO < 0.25 g m
3) there is a failure in nitritation efficiency,
and there is no conversion of SNHx to SNO2 (Fig. 6b and f). At higher
DO values, XNOB have a competitive advantage compared to XANX .

Fig. 5. Intra-granule model prediction for option A2 when PRE is active (A2,1 ¼ PN/ANX þ pre-treatment) or not active (A2,2 ¼ PN/ANX without pre-treatment). In the plot,
0 represents the centre of the granule and 0.5 the surface of the granule that is in contact with the liquid phase.
 X. Flores-Alsina et al. / Water Research 156 (2019) 264e276
Fig. 6. Local (DO set-point and TSSop in REAC) and Global/plant-wide (Erecovery in UASB and DO set-point in REAC) modifications of the operational settings (1st and 2nd row,
respectively) and their effects on TKN, TN, Eaeration and Enet for option A2 (PRE þ PN/ANX). In this case A2 includes PN/ANX þ pre-treatment.
As a result, there is a decrease in the overall N removal efficiency
due to the substantial amount of SNO3 that is formed in the aqueous
phase. This can be compensated by increasing the TSSop (and
thereby the ratio of Vbiofilm to Vreactor) or increasing the quantity of
organics coming from the digester. Nevertheless, this practice has a
negative effect on the entire energy balance in the plant. (Eaeration,
Erecovery and Enet).
3.3.2. Importance of plant-wide approaches to avoid suboptimal
outcomes
The potential N removal and Enet maximization in REAC by
modifying TSSop and DO setpoint is a clear example of sub-
optimization. For example, it was possible to get better Enet
values when the UASB and REAC where modified together (see
Fig. 6, second row) instead of making local changes in REAC (up to
155 MWh.d
1 compared to 148 MWh.d
1 while N removal effi-
ciency is maintained). The mass and energy balances provided by
the PWM can be very useful to suggest ways of better managing the
different industrial streams within the factory, and achieving better
operational process solutions. In addition, the optimization exam-
ples depicted in Fig. 6 could be combined with other approaches,
such as pH and sulfide production control in the AD. In this regard,
previous studies (Feldman et al., 2018) have demonstrated that a
good pH control could reduce the operational cost of the digester by
up to 45% without risking the overall process performance. Another
interesting example is reported in Irizar et al. (2018), where
advanced software sensors are applied to a similar type of UASB
reactor to ensure stable operating conditions.
4. Discussion
4.1. Evaluation of (realistic) control strategies in biofilm systems
The PN/ANX model developed during this study will allow
multiple possibilities, thus opening a new field of research. As
273
mentioned in the introduction, previous studies including biofilm
reactors did not include sophisticated control elements mainly due
to software limitations of the platform on which these simulations
were executed (Reichert, 2001). The biofilm model presented here,
while carrying a high development overhead, offers a number of
advantages over classic lumped parameter implementations
(which normally rely on increased solids retention time, Batstone
et al., 2002). This includes inherent incorporation of diffusional
limitations, consideration of how specific gaps will be selectively
retained, and in-biofilm interactions. Many of these factors are
quite important to short term dynamics, which in turn, impact
control strategies. The implementation presented in this paper has
shown that the computational demand is manageable. This
implementation (which will be freely distributed) will allow users
to develop/simulate control strategies using the most common on-
line sensors, such as pH, DO, NH4, NO3, NO2 and conductivity
(Lackner et al., 2014). Implementation of pH has been especially
challenging due to the numerical problems that this variable causes
(Batstone et al., 2012). Some of these implemented sensors might
have considerable noise, response time and delay (Rieger et al.,
2003). The same applies for actuators. Therefore, such elements
should be included in order to evaluate controllers realistically
(Rosen et al., 2008). The special numerical routines implemented in
these models allowed simulating these elements without problems
(Flores-Alsina et al., 2015; Feldman et al., 2017).
4.2. Generation of alternatives
The options evaluated in this case study have been systemati-
cally generated, but they are based on general process knowledge.
The combination of plant-wide models (Ekama, 2009; Barat et al.,
2013; Solon et al., 2017, Ferna ndez-Are valo et al., 2017) with envi-
ronmental decision support systems (EDSS) for suggesting process
flow sheets (Poch et al., 2004; Garrido-Baserba et al., 2012; Castillo
et al., 2016) is a promising line of research. Indeed, the combination
274 X. Flores-Alsina et al. / Water Research 156 (2019) 264e276
of process heuristics with mixed integer non-linear programming
could yield a good set of technologies to be further evaluated using
dynamic process models. There is a bright future if both tools go
hand in hand when retrofitting an existing plant during the tran-
sition from WWTPs to Water Resource Recovery Facilities (Daigger,
2011).
4.3. Outlook, limitations and future research directions
It is important to highlight that only the ADM section of the in-
tegrated approach presented in this study has been validated
experimentally with full-scale data. Process equations and kinetic
and stoichiometric parameters for the ASM part have been extracted
from literature based on validated case studies. The N removal part
using a granular system has been tested in Vangsgaard et al. (2013).
Methane and sulfide process equations were taken from previous
studies dealing with a similar type of wastewater (Winkler et al.,
2014; Chen et al., 2016a, b). pH predictions and precipitation equa-
tions are validated in Kazadi-Mbamba et al. (2015a, b). Therefore, the
results from the simulation study seem to be correct and reliable,
within the normal uncertainty range of simulation studies of
wastewater treatment systems. It is interesting to note that, as a
result of this study, the company involved will implement alterna-
tive PRE þ PN/ANX at full scale as a post-digestion treatment tech-
nology. The potential better results in Enet were the key issue to tip
the scale towards selection of this option.
Last but not least, the reader should be aware that the PWM
presented in this case study shows a good compromise between
model complexity and prediction capabilities. Nevertheless, the list
of relevant processes is far from being complete. PN processes (or
systems in general where there is NO
2 accumulation) are well-
known for producing relatively high quantities of N2O emissions.
N2O is a powerful greenhouse gas that is 300 times stronger than CO2
(IPCC, 2013). CH4 stripping is an important contribution to green-
house gas formation as well. As a consequence, a more integrated
assessment should include a compromise between effluent quality,
operational cost and greenhouse gas emissions (Flores-Alsina et al.,
2014; Ni and Yuan 2015). Other important factors not included in
this study when evaluating PN/ANX systems is the variation of the
size of the granular sludge (Volcke et al., 2010, 2012) and the effects
of intra-granular precipitation (Johansson et al., 2017; Feldman et al.,
2018). Sulfide inhibition should also be better represented in the
model. In this study, an arbitrarily high value was defined to account
for the S effect on biomass growth. Nevertheless, there are numerous
studies that report quite complex interactions between XANX and
XSRB (Jin et al., 2012, 2013). The present study could be com-
plemented with more thorough energy balances (Ferna ndez-
Arevalo et al., 2017; Longo et al 2018) or life cycle assessment
(Corominas et al., 2013). Uncertainty/risk analysis, a central concept
for model-based evaluation of alternatives (Belia et al., 2009), is an
important issue that should be considered as well. The consideration
of uncertainty/risk factors within the evaluation procedure can
substantially change the evaluation results, and preliminary
selected technologies might come up as less desirable while other
options retain a higher chance of success under uncertainty.
5. Conclusions
The main findings of this study can be summarized in the
following points:
 A plant wide model (PWM) was presented, consisting of a UASB
reactor based on an ADM type model þ two post-treatment
options based on an ASM type model. Special features to inter-
face both models had to be developed. The ASM approach was
implemented in two reactor designs (activated sludge/aerobic
granular sludge). Sensor and actuator models were included to
avoid unrealistic control actions and allow future control pos-
sibilities. The PWM was simulated and the performance of the
different flow diagrams evaluated in terms of pollution removal
(nitrogen, CH4, H2S), energy production (Erecovery), energy con-
sumption (Eaeration/Epumping) and net energy balance (Enet).
 Model-based evaluation revealed that PN/ANX (option A2)
demanded lower aeration rates (Eaeration ¼ 1.75 kWh/kg Nre-
moved) and potentially allowed more energy recovery in the
UASB reactor (Erecovery ¼ up to 156.7 MWh.d
1) since the
Anammox bacteria in charge of denitrification did not need
organics as electron donor. As a result, the Enet
(¼153.9 MWh.d
1) was potentially higher. Conventional N/DN
(option A1) as post-treatment of UASB effluent required higher
aeration rates (Eaeration/Epumping ¼ 3.95 kWh/kg Nremoved) to
convert the NHþ 4 all the way to NO3, and lower conversion in the
AD (Erecovery is down to 132.8 MWh.d
1) to allow denitrification,
i.e. resulting in a considerably lower net energy production Enet
(¼127.1 MWh.d
1). Calculations show that 95 kWh.d
1 is lost in
terms of energy production per extra kg N.d
1 removed. These
results are in agreement with the well-known advantages of PN/
ANX process over the conventional nitrification/denitrification
process.
 The addition of an aeration pre-treatment tank (PRE) before the
PN/ANX reactor increases COD/H2S oxidation, CH4/CO2 strip-
ping, Ca precipitation and increases pH. These processes change
the wastewater influent composition, and due to the presence of
XDAMO and XSRB this ends up changing the granule microbial
distribution. pH variation and risk of precipitation are also
predicted by the model.
 The modification of the UASB conversion efficiency þ DO set-
point in the PN/ANX reactor rendered better results (Enet) than
the local (TSSop þ DO set-point) optimization of the PN/ANX
operational settings. In this way, it was possible to take full
advantage of the interactions amongst the different units and
hence avoid sub-optimal outcomes.
 The study has shown the potential of using simulations for
virtually evaluating competing options before full-scale imple-
mentation. Relevant information about the general mass and
energy balances will yield additional information that will help
the process engineer to take the right decision in terms of
treatment option to be selected.
6. Software availability
The Matlab/Simulink implementation of the UASB reactor
model, the N/DN reactor model and the PN/ANX reactor model plus
model interfaces are available on request. The code is fully docu-
mented and presented results can be reproduced. To express an
interest, please contact Dr. Xavier Flores-Alsina (xfa@kt.dtu.dk) or
Professor Krist V. Gernaey (kvg@kt.dtu.dk) at the Department of
Chemical and Biochemical Engineering at the Technical University
of Denmark.
Appendix A. Supplementary data
Supplementary data to this article can be found online at
https://doi.org/10.1016/j.watres.2019.02.035.
References
APHA, 2012. Standard Methods for the Examination of Water and Wastewater.
American Public Health Association, Washington, D.C., USA.
 X. Flores-Alsina et al. / Water Research 156 (2019) 264e276
Barat, R., Serralta, J., Ruano, V., Jimenez, E., Ribes, J., Seco, A., Ferrer, J., 2013. Bio-
logical Nutrient Removal no 2 (BNRM2): a general model for wastewater
treatment plants. Water Sci. Technol. 67 (7), 1481e1489.
Barker, P.S., Dold, P.L., 1997. General model for biological nutrient removal activated-
sludge systems: model presentation. Water Environ. Res. 69 (5), 969e984.
Batstone, D.J., Keller, J., Angelidaki, I., Kalyuzhnyi, S.V., Pavlostathis, S.G., Rozzi, A.,
Sanders, W.T.M., Siegrist, H., Vavilin, V.A., 2002. Anaerobic Digestion Model No.
1. IWA Scientific and Technical Report No. 13. IWA Publishing, London, UK.
Batstone, D.J., Puyol, D., Flores-Alsina, X., Rodriguez, J., 2015. Mathematical
modelling of anaerobic digestion processes: applications and future needs. Rev.
Environ. Sci. Biotechnol. 14 (4), 595e613.
Batstone, D.J., Torrijos, M.J., Ruiz, C., Schmidt, J.E., 2004. Use of an anaerobic
sequencing batch reactor for parameter estimation in modelling of anaerobic
digestion. Water Sci. Technol. 50 (10), 295e303.
Batstone, D.J., Amerlinck, Y., Ekama, G., Goel, R., Grau, P., Johnson, B., Kaya, I.,
Steyer, J.-P., Tait, S., Tak
acs, I., Vanrolleghem, P.A., Brouckaert, C.J., Volcke, E.I.P.,
2012. Towards a generalized physicochemical framework. Water Sci. Technol.
66 (6), 1147e1161.
Belia, E., Amerlinck, Y., Benedetti, L., Johnson, B., Sin, G., Vanrolleghem, P.A.,
Gernaey, K.V., Gillot, S., Neuman, M., Rieger, L., Shaw, A., Villez, K., 2009.
Wastewater treatment modelling: dealing with uncertainties. Water Sci.
Technol. 60 (8), 1929e1941.
Benedetti, L., De Keyser, W., Nopens, I., Vanrolleghem, P.A., 2010. Probabilistic
modelling and evaluation of wastewater treatment plant upgrades in the EU
Water Framework Directive context. J. Hydroinf. 12 (4), 380e395.
Cao, Y., van Loosdrecht, M.C.M., Daigger, G.T., 2017. Mainstream partial nitritation-
anammox in municipal wastewater treatment: status, bottlenecks, and
further studies. Appl. Microbiol. Biotechnol. 101 (4), 1365e1383.
Castillo, A., Cheali, P., Go mez, V., Comas, J., Poch, M., Sin, G., 2016. An integrated
knowledge based and optimization tool for the sustainable selection of
wastewater treatment process concepts. Environ. Model. Softw 84, 177e192.
Castro-Barros, C.M., Ho, L.T., Winkler, M.K.H., Volcke, E.I.P., 2018. Integration of
methane removal in aerobic anammox-based granular sludge reactors. Environ.
Technol. 39 (13), 1615e1625.
Chen, X., Guo, J., Xie, G.J., Liu, Y., Yuan, Z., Ni, B.J., 2016a. A new approach to
simultaneous ammonium and dissolved methane removal from anaerobic
digestion liquor: a model-based investigation of feasibility. Water Res. 85,
295e303, 2015.
Chen, X., Liu, Y., Peng, L., Yuan, Z., Ni, B.J., 2016b. Model-based feasibility assessment
of membrane biofilm reactor to achieve simultaneous ammonium, dissolved
methane, and sulfide removal from anaerobic digestion liquor. Sci. Rep. 6.
Article number: 25114.
Copp, J.B. (Ed.), 2002. The COST Simulation Benchmark: Description and Simulator
Manual. Office for Official Publications of the European Community,
Luxembourg, ISBN 92-894-1658-0.
Corominas, L., Foley, J., Guest, J.S., Hospido, A., Larsen, H.F., Morera, S., Shaw, A.,
2013. Life cycle assessment applied to wastewater treatment: state of the art.
Water Res. 47 (15), 5480e5492.
Daelman, M.R.J., Van Eynde, T., van Loosdrecht, M.C.M., Volcke, E.I.P., 2014. Effect of
process design and operating parameters on aerobic methane oxidation in
municipal WWTPs. Water Res. 66, 308e319.
Daigger, G.T., 2011. Changing paradigms: from wastewater treatment to resource
recovery. Proc. Water Environ. Fed. 2011 (6), 942e957.
Ekama, G.A., 2009. Using bioprocess stoichiometry to build a plant-wide mass
balance based steady-state WWTP model. Water Res. 43 (8), 2101e2120.
Feldman, H., Flores-Alsina, X., Ramin, P., Kjellberg, K., Batstone, D.J., Jeppsson, U.,
Gernaey, K.V., 2017. Modelling an industrial anaerobic granular reactor using a
multi-scale approach. Water Res. 126, 488e500.
Feldman, H., Flores-Alsina, X., Ramin, P., Kjellberg, K., Jeppsson, U., Batstone, D.J.,
Gernaey, K.V., 2018. Model-based analysis and optimization of a full-scale in-
dustrial high rate anaerobic bioreactor. Biotechnol. Bioeng. 115 (11), 2726e2739.
Fernandez-Are valo, T., Lizarralde, I., Fdz-Polanco, F., Perez-Elvira, S.I., Garrido, J.M.,
Puig, S., Poch, M., Grau, P., Ayesa, E., 2017. Quantitative assessment of energy
and resource recovery in wastewater treatment plants based on plant-wide
simulations. Water Res. 118, 272e288.
Flores-Alsina, X., Corominas, L., Neumann, M.B., Vanrolleghem, P.A., 2012. Assessing
the use of activated sludge processes design guidelines in wastewater treat-
ment plant projects. Environ. Model. Softw 38, 50e58.
Flores-Alsina, X., Kazadi Mbamba, C., Solon, K., Vrecko, D., Tait, S., Batstone, D.J.,
Jeppsson, U., Gernaey, K.V., 2015. A plant-wide aqueous phase chemistry
module describing pH variations and ion speciation/pairing in wastewater
treatment models. Water Res. 85, 255e265.
Flores-Alsina, X., Solon, K., Kazadi Mbamba, C., Tait, S., Jeppsson, U., Gernaey, K.V.,
Batstone, D.J., 2016. Modelling phosphorus, sulphur and iron interactions dur-
ing the dynamic simulation of anaerobic digestion processes. Water Res. 95,
370e382.
Flores-Alsina, X., Arnell, M., Amerlinck, Y., Corominas, L., Gernaey, K.V., Guo, L.,
Lindblom, E., Nopens, I., Porro, J., Shaw, A., Snip, L., Vanrolleghem, P.A.,
Jeppsson, U., 2014. Balancing effluent quality, economical cost and greenhouse
gas emissions during the evaluation of plant-wide wastewater treatment plant
control strategies. Sci. Total Environ. 466e467, 616e624.
Garrido-Baserba, M., Reif, R., Herna ndez, F., Poch, M., 2012. Implementation of a
knowledge-based methodology in a decision support system for the design of
suitable wastewater treatment process flow diagrams. J. Environ. Manag. 112,
384e391.
275
Benchmarking of control strategies for wastewater treatment plants. In:
Gernaey, K.V., Jeppsson, U., Vanrolleghem, P.A., Copp, J.B. (Eds.), 2014. IWA
Scientific and Technical Report No. 23. IWA Publishing, London, UK.
Grau, P., Copp, J., Vanrolleghem, P.A., Tak acs, I., Ayesa, E., 2009. A comparative
analysis of different approaches for integrated WWTP modelling. Water Sci.
Technol. 59 (1), 141e147.
Grau, P., de Gracia, M., Vanrolleghem, P.A., Ayesa, E., 2007. A new plant-wide
modelling methodology for WWTPs. Water Res. 41 (19), 4357e4372.
Gutierrez, O., Park, D., Sharma, K.R., Yuan, Z., 2010. Iron salts dosage for sulfide
control in sewers induces chemical phosphorus removal during wastewater
treatment. Water Res. 44 (11), 3467e3475.
Henze, M., Gujer, W., Mino, T., van Loosdrecht, M.C.M., 2000. Activated Sludge
Models ASM1, ASM2, ASM2d, and ASM3. IWA Scientific and Technical Report
No. 9. IWA Publishing, London, UK.
Hug, T., Benedetti, L., Hall, E.R., Johnson, B.R., Morgenroth, E., Nopens, I., Rieger, L.,
Shaw, A., Vanrolleghem, P.A., 2009. Wastewater treatment models in teaching
and training: the mismatch between education and requirements for jobs.
Water Sci. Technol. 59 (4), 745e753.
IPCC, 2013. Climate Change 2013: the Physical Science Basis. Working Group I
Contribution to the IPCC 5th Assessment Report - Changes to the Underlying
Scientific/technical Assessment. IPCC, Cambridge, United Kingdom and New
York, NY, USA.
Irizar, I., Roche, E., Beltra n, S., Aymerich, E., Esteban-Gutie rrez, M., 2018. Model-
based design of a software sensor for real-time diagnosis of the stability con-
ditions in high-rate anaerobic reactors e full-scale application to internal cir-
culation technology. Water Res. 143, 479e491.
Jeppsson, U., Alex, J., Batstone, D., Benedetti, L., Comas, J., Copp, J.B., Corominas, L.,
Flores-Alsina, X., Gernaey, K.V., Nopens, I., Pons, M.-N., Rodriguez-Roda, I.,
Rosen, C., Steyer, J.-P., Vanrolleghem, P.A., Volcke, E.I.P., Vrecko, D., 2013.
Benchmark simulation models: quo vadis? Water Sci. Technol. 68 (1), 1e15.
Jeppsson, U., Pons, M.N., Nopens, I., Alex, J., Copp, J.B., Gernaey, K.V., Rosen, C.,
Steyer, J.P., Vanrolleghem, P.A., 2007. Benchmark Simulation Model No 2 e
general protocol and exploratory case studies. Water Sci. Technol. 56 (8),
287e295.
Jin, R.J., Yang, G.F., Zhang, Q.Q., Ma, C., Yu, J.J., Xing, B.S., 2013. The effect of sulfide
inhibition on the ANAMMOX process. Water Res. 47, 1459e1469.
Jin, R.J., Yang, G.F., Yu, J.J., Zheng, P., 2012. The inhibition of the Anammox process: a
review. Chem. Eng. J. 197, 67e79.
Johansson, S., Ruscalleda, M., Colprim, J., 2017. Phosphorus recovery through bio-
logically induced precipitation by partial nitritation-anammox granular
biomass. Chem. Eng. J. 327, 881e888.
Kalyuzhnyi, S.V., Fedorovich, V.V., 1998. Mathematical modelling of competition
between sulphate reduction and methanogenesis in anaerobic reactors. Bio-
resour. Technol. 65, 227e242.
Kazadi Mbama, C., Flores-Alsina, X., Batstone, D.J., Tait, S., 2016. Validation of a
plant-wide modelling approach with minerals precipitation in a full-scale
WWTP. Water Res. 100, 169e183.
Kazadi Mbamaba, C., Flores-Alsina, X., Batstone, D.J., Tait, S., 2015b. A systematic
study of multiple minerals precipitation modelling in wastewater treatment.
Water Res. 85, 359e370.
Kazadi Mbamba, C., Flores-Alsina, X., Batstone, D.J., Tait, S., 2015a. A generalised
chemical precipitation modelling approach in wastewater treatment applied to
calcite. Water Res. 68, 342e353.
Lackner, S., Gilberta, E.M., Siegfried, E.V., Joss, A., Horna, H., van Loosdrecht, M.C.M.,
2014. Full-scale partial nitritation/anammox experiences e an application sur-
vey. Water Res. 55, 292e303.
Lackner, S., Terada, A., Smets, B.F., 2008. Heterotrophic activity compromises
autotrophic nitrogen removal in membrane-aerated biofilms: results of a
modeling study. Water Res. 42 (4e5), 1102e1112.
Lemaire, R., Webb, R.I., Yuan, Z., 2009. Micro-scale observations of the structure of
aerobic microbial granules used for the treatment of nutrient-rich industrial
wastewater. ISME J. 2 (5), 528e541.
Longo, S., Hospido, A., Lema, J.M., Maurico-Iglesias, M., 2018. A systematic meth-
odology for the robust quantification of energy efficiency at wastewater
treatment plants featuring Data Envelopment Analysis. Water Res. 141,
317e328.
Musvoto, E.V., Wentzel, M.C., Ekama, G.A., 2000. Integrated chemical-physical
processes modelling e II. Simulating aeration treatment of anaerobic digester
supernatants. Water Res. 34 (6), 1868e1880.
Ni, B.J., Chen, Y., Liu, S., Fang, F., Xie, W., Yu, H., 2009. Modeling a granule-based
anaerobic ammonium oxidizing (Anammox) process. Biotechnol. Bioeng. 103
(3), 490e499.
Ni, B.J., Yuan, Z., 2015. Recent advances in mathematical modeling of nitrous oxides
emissions from wastewater treatment processes. Water Res. 87, 336e346.
Nopens, I., Batstone, D.J., Copp, J.B., Jeppsson, U., Volcke, E., Alex, J.,
Vanrolleghem, P.A., 2009. An ASM/ADM model interface for dynamic plant-
wide simulation. Water Res. 43 (7), 1913e1923.
Perez, J., Lotti, T., Kleerebezem, R., Picioreanu, C., van Loosdrecht, M.C.M., 2014.
Outcompeting nitrite-oxidizing bacteria in single-stage nitrogen removal in
sewage treatment plants: a model-based study. Water Res. 66, 208e218.
Poch, M., Comas, J., Rodríguez-Roda, I., Sa nchez-Marre , M., Cortes, U., 2004.
Designing and building real environmental decision support systems. Environ.
Model. Softw 19, 857e873.
Poot, V., Hoekstra, M., Geleijnse, M.A.A., van Loosdrecht, M.C.M., Pe rez, J., 2016.
Effects of the residual ammonium concentration on NOB repression during
276 X. Flores-Alsina et al. / Water Research 156 (2019) 264e276
partial nitritation with granular sludge. Water Res. 106, 518e530.
Press, H., Teukolsky, S.A., Vetterling, W.T., Flannery, B.P., 2007. Numerical Recipes:
the Art of Scientific Computing, third ed. Cambridge University Press, New York,
NY, USA.
Reichert, P., 2001. AQUASIM 2.0-Computer Program for the Identification and
Simulation of Aquatic Systems. Swiss Federal Institute for Environmental Sci-
ence and Technology (EAWAG), Dubendorf, Switzerland.
Rieger, L., Alex, J., Winkler, S., Boehler, M., Thomann, M., Siegrist, H., 2003. Progress
in sensor technology - progress in process control Part I: sensor property
investigation and classification. Water Sci. Technol. 47 (2), 103e112.
Rivas, A., Irizar, I., Ayesa, E., 2008. Model-based optimisation of wastewater treat-
ment plants design. Environ. Model. Softw 23 (4), 435e450.
Rodriguez-Roda, I., Sa nchez-Marre , M., Comas, J., Baeza, J., Colprim, J., Lafuente, J.,
Cortes, U., Poch, M., 2002. A hybrid supervisory system to support WWTP
operation: implementation and validation. Water Sci. Technol. 45 (4e5),
289e297.
Rosen, C., Vrecko, D., Gernaey, K.V., Pons, M.-N., Jeppsson, U., 2006. Implementing
ADM1 for plant-wide benchmark simulations in Matlab/Simulink. Water Sci.
Technol. 54 (4), 11e19.
Rosen, C., Jeppsson, U., Rieger, L., Vanrolleghem, P.A., 2008. Adding realism to
simulated sensors and actuators. Water Sci. Technol. 57 (3), 337e344.
Saravanan, V., Sreekrishnan, T.R., 2006. Modelling anaerobic biofilm reactors e a
review. J. Environ. Manag. 81, 1e18.
Skogestad, S., 2000. Plantwide control: the search for the self-optimizing control
structure. J. Process Control 10, 487e507.
Solon, K., Flores-Alsina, X., Kazadi Mbamba, C., Volcke, E.I.P., Tait, S., Batstone, D.J.,
Gernaey, K.V., Jeppsson, U., 2015. Effects of ionic strength and ion pairing on
(plant-wide) modelling of anaerobic digestion processes. Water Res. 70,
235e245.
Solon, K., Flores-Alsina, X., Kazadi-Mbamba, C., Ikumi, D., Volcke, E.I.P.,
Vaneeckhaute, C., Ekama, G., Vanrolleghem, P.A., Batstone, D.J., Gernaey, K.V.,
Jeppsson, U., 2017. Plant-wide modelling of phosphorus transformations in
wastewater treatment systems: impacts of control and operational strategies.
Water Res. 113, 97e110.
Stumm, W., Morgan, J.J., 1996. In: Schnoor, J.L., Zehnder, A. (Eds.), Aquatic Chem-
istry: Chemical Equilibria and Rates in Natural Waters. John Wiley and Sons,
New York, NY, USA.
Tak
acs, I., Patry, G.G., Nolasco, D., 1991. A dynamic model of the clarification
thickening process. Water Res. 25 (10), 1263e1271.
Tchobanoglous, G., Burton, F.L., Stensel, H.D., 2003. Wastewater Engineering:
Treatment and Reuse, fourth ed. McGraw-Hill Education, Massachusetts, USA.
Van Lier, J., 2015. Celebrating 40 years anaerobic sludge bed reactors for industrial
wastewater treatment. Rev. Environ. Sci. Biotechnol. 14 (14), 681e702.
Vangsgaard, A.K., Mutlu, G., Gernaey, K.V., Smets, B.F., Sin, G., 2013. Calibration and
validation of a model describing complete autotrophic nitrogen removal in a
granular SBR system. J. Chem. Technol. Biotechnol. 88, 2007e2015.
Vangsgaard, A.K., Mauricio-Iglesias, M., Gernaey, K.V., Smets, B.F., Sin, G., 2012.
Sensitivity analysis of autotrophic N removal by a granule-based bioreactor:
influence of mass transfer versus microbial kinetics. Bioresour. Technol. 123,
230e241.
Vela, J.D., Stadler, L.B., Martin, K.J., Raskin, L., Bott, C.B., Love, N.G., 2015. Prospects
for biological nitrogen removal from anaerobic effluents during mainstream
wastewater treatment. Environ. Sci. Technol. Lett. 2 (9), 234e244.
Volcke, E.I.P., Picioreanu, C., De Baets, B., van Loosdrecht, M.C.M., 2010. Effect of
granule size on autotrophic nitrogen removal in a granular sludge reactor.
Environ. Technol. 31 (11), 1271e1280.
Volcke, E.I.P., Picioreanu, C., De Baets, B., van Loosdrecht, M.C.M., 2012. The granule
size distribution in an Anammox-based granular sludge reactor affects the
conversion e implications for modeling. Biotechnol. Bioeng. 109 (7),
1629e1636.
Volcke, E.I.P., van Loosdrecht, M.C.M., Vanrolleghem, P.A., 2006a. Continuity-based
model interfacing for plant-wide simulation: a general approach. Water Res. 40
(15), 2817e2828.
Volcke, E.I.P., Gernaey, K.V., Vrecko, D., Jeppsson, U., van Loosdrecht, M.C.M.,
Vanrolleghem, P.A., 2006b. Plant-wide (BSM2) evaluation of reject water
treatment with a SHARON-Anammox process. Water Sci. Technol. 54 (8),
93e100.
Wanner, O., Eberl, H.J., Morgenroth, E., Noguera, D.R., Picioreanu, C., Rittmann, B.E.,
van Loosdrecht, M.C.M., 2006. Mathematical Modeling of Biofilms. In: IWA
Scientific and Technical Report No. 18. IWA Publishing, London, UK.
Winkler, M.K., Ettwig, K.F., Vannecke, T.P., Stultiens, K., Bogdan, A., Kartal, B.,
Volcke, E.I., 2014. Modelling simultaneous anaerobic methane and ammonium
removal in a granular sludge reactor. Water Res. 73, 323-231.
Zaher, U., Grau, P., Benedetti, L., Ayesa, E., Vanrolleghem, P.A., 2007. Transformers for
interfacing anaerobic digestion models to pre- and post-treatment processes in
a plant-wide modelling context. Environ. Model. Softw 22 (1), 40e58.
Zhang, L., De Schryver, P., De Gusseme, B., De Muynck, W., Boon, N., Verstraete, W.,
2008. Chemical and biological technologies for hydrogen sulfide emission
control in sewer systems: a review. Water Res. 42 (1e2), 1e12.