Water Research 215 (2022) 118223 Contents lists available at ScienceDirect Water Research journal homepage: www.elsevier.com/locate/watres A plant-wide model describing GHG emissions and nutrient recovery options for water resource recovery facilities Borja Solís a, Albert Guisasola a, *, Xavier Flores-Alsina b, Ulf Jeppsson c, Juan Antonio Baeza a a GENOCOV, Departament d’Enginyeria Química, Biològica i Ambiental, Escola d’Enginyeria, Universitat Autònoma de Barcelona, Cerdanyola del Vallès, 08193 Barcelona, Spain b PROSYS Research Centre, Department of Chemical and Biochemical Engineering, Technical University of Denmark, Building 229, DK-2800 Kgs, Lyngby, Denmark c Division of Industrial Electrical Engineering and Automation (IEA), Department of Biomedical Engineering, Lund University, Box 118, SE-221 00 Lund, Sweden A R T I C L E I N F O A B S T R A C T Keywords: In this study, a plant-wide model describing the fate of C, N and P compounds, upgraded to account for (on-site/ Benchmarking off-site) greenhouse gas (GHG) emissions, was implemented within the International Water Association (IWA) Control strategies Benchmarking Simulation Model No. 2 (BSM2) framework. The proposed approach includes the main biological EBPR N2O production pathways and mechanistically describes CO2 (biogenic/non-biogenic) emissions in the activated GHG emissions BSM2 sludge reactors as well as the biogas production (CO2/CH4) from the anaerobic digester. Indirect GHG emissions for power generation, chemical usage, effluent disposal and sludge storage and reuse are also included using static factors for CO2, CH4 and N2O. Global and individual mass balances were quantified to investigate the fluxes of the different components. Novel strategies, such as the combination of different cascade controllers in the biological reactors and struvite precipitation in the sludge line, were proposed in order to obtain high plant performance as well as nutrient recovery and mitigation of the GHG emissions in a plant-wide context. The implemented control strategies led to an overall more sustainable and efficient plant performance in terms of better effluent quality, reduced operational cost and lower GHG emissions. The lowest N2O and overall GHG emissions were achieved when ammonium and soluble nitrous oxide in the aerobic reactors were controlled and struvite was recovered in the reject water stream, achieving a reduction of 27% for N2O and 9% for total GHG, compared to the open loop configuration. simulating WRRF under different scenarios and of designing novel control strategies for a better performance. 1. Introduction Plant-wide WRRF modelling comprises chemical and physico- chemical models to assess the new challenges of the wastewater treat­ In recent years, the scarcity of natural resources and the concern ment. Particularly, precipitation models of common chemical com­ about climate change have shifted the water sector paradigm. Therefore, pounds in wastewater (Hauduc et al., 2015; Kazadi Mbamba et al., 2016, wastewater treatment plants (WWTPs) are becoming water resource 2015b), aqueous phase chemistry models (Flores-Alsina et al., 2015; recovery facilities (WRRFs). This fact has promoted both the chemical Solon et al., 2015) and mass transfer models (Amaral et al., 2019; Liz­ and environmental engineering community and the water industry to arralde et al., 2015) have been proposed and calibrated to study widen the scope of these utilities. To better understand and to design different subprocesses of a WWRF in addition to the traditional activated these new facilities, plant-wide modelling tools have become essential sludge models (ASM) based on biological processes (Batstone et al., (Jeppsson et al., 2013). Wastewater treatment modelling researchers 2002; Henze et al., 2015). Recently, Solon and co-workers (Solon et al., have integrated the main unit operations of a WRRF (primary clarifier, 2017) proposed a novel plant-wide model capable of predicting the fate biological reactor, secondary settler, thickeners, anaerobic digester, of phosphorus (P) in both water and sludge lines as well as the in­ dewatering unit, etc.) to account for all the interactions amongst pro­ teractions with sulphur (S) and iron (Fe) thanks to the implementation of cesses (Barat et al., 2013; Gernaey et al., 2014; Grau et al., 2007; Hau­ comprehensive physico-chemical process models. This work combined a duc et al., 2019; Solon et al., 2017; Vaneeckhaute et al., 2018) in view of * Corresponding author. E-mail addresses: borja.solis@uab.cat (B. Solís), albert.guisasola@uab.cat (A. Guisasola), xfa@kt.dtu.dk (X. Flores-Alsina), ulf.jeppsson@iea.lth.se (U. Jeppsson), JuanAntonio.Baeza@uab.cat (J.A. Baeza). https://doi.org/10.1016/j.watres.2022.118223 Received 8 September 2021; Received in revised form 20 January 2022; Accepted 23 February 2022 Available online 25 February 2022 0043-1354/© 2022 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). B. Solís et al. Water Research 215 (2022) 118223 Nomenclature ND pathway Nitrifier Denitrification pathway NH2OH Hydroxylamine A2/O Anaerobic-Anoxic-Aerobic NN pathway Nitrifier Nitrification pathway ABAC Aeration-based ammonium controller NO Nitric Oxide AD Anaerobic digester NO2− Nitrite ADM1 Anaerobic Digestion Model No. 1 NO3− Nitrate AER Aerobic section NOx Oxidized forms of nitrogen ANAER Anaerobic section O2 Oxygen gas ANOX Anoxic section OCI Operational Cost Index AOB Ammonia Oxidizing Bacteria OHO Ordinary Heterotrophic Organism AS Activated Sludge P Phosphorus ASM Activated Sludge Model PAO Polyphosphate Accumulating Organisms ASM2d Activated Sludge Model No. 2d PCM Physico-Chemical Models BOD Biological Oxygen Demand Pi Partial pressure of the gas specie i BSM2 Benchmark Simulation Model No. 2 PI Proportional Integral controller BSM2-PSFe BSM2 for Phosphorus, Sulfur and Iron Pinorg Inorganic phosphorus BSM2G BSM2 Greenhouse gas Porg Organic phosphorus C Carbon PRIM Primary clarifier Ca3(PO4)2 Amorphous calcium phosphate PROCESSAS-AD Process ASM-ADM interface Ca5(PO4)3OH Hydroxyapatite Qw Purge flow rate CaCO3 Calcite S Sulphur CBIM Continuity-based interfacing method SA Acetate (ASM) (gCOD m− 3) CH4 Methane Saa Amino acids (ADM) (kgCOD m− 3) CO2 Carbon dioxide Sac Acetate (ADM) (kgCOD m− 3) CO2e Carbon dioxide equivalents SEC Secondary settler COD Chemical Oxygen Demand SF Soluble fermentable (ASM) (gCOD m− 3) CONVAS-AD Conversion ASM-ADM interface Sfa Fatty acids (ADM) (kgCOD m− 3) DEN pathway Denitrification pathway SI Saturation index for mineral precipitation DEW Dewatering unit SI Unbiodegradable soluble organics (ASM) (gCOD m− 3) Di Diffusivity of component i in liquid phase (m2 d− 1) SIC Inorganic Carbon (ASM, ADM) (kmol m− 3) DO Dissolved Oxygen SN2O Nitrous oxide (ASM) (gN m− 3) DO2 Diffusivity of oxygen in liquid phase (m2 d− 1) SNH2OH Hydroxylamine (ASM) (gN m− 3) EBPR Enhanced Biological Phosphorus Removal SNH4 Ammonium plus ammonia nitrogen (ASM) (gN m− 3) N2O-EF Nitrous Oxide Emission Factor (%) SNO Nitric oxide (ASM) (gN m− 3) Eproduction Electricity production SNO2 Nitrite (ASM) (gN m− 3) EQI Effluent Quality Index SO2 Dissolved oxygen (ASM) (gO m− 3) FA Free Ammonia SPO4 Phosphate (ASM) (gP m− 3) Fe Iron SRT Sludge Retention Time FeCl3 Ferric Chloride Ssu Sugars (ADM) (kgCOD m− 3) FeS Iron Sulphide ST Storage tank FNA Free Nitrous Acid T Temperature GHG Greenhouse gas THK Thickener GWP Global Warming Potential TIV Time In Violation H2 Hydrogen gas TKN Total Kjeldahl Nitrogen H2O Water TN Total Nitrogen H2S Hydrogen sulphide TP Total Phosphorus HCO3− Bicarbonate TSS Total Suspended Solids HFO Hydrous Ferric Oxides VS Volatile Solids IC Inorganic Carbon WRRF Water Resource Recovery Facility KH,i Henry’s constant for the specie i WWTP Waste Water Treatment Plant kLai Mass transfer coefficient for component i (d− 1) XAOB Ammonia Oxidizing Bacteria (ASM) (gCOD m− 3) kLaO2 Mass transfer coefficient for oxygen (d− 1) Xch Carbohydrates (ADM) (kgCOD m− 3) KNH4PO4 k-Struvite Xi Mineral concentration in solid phase (PCM) (kmol m− 3) Mg(OH)2 Magnesium Hydroxide XI Inert particulates organics (ASM, ADM) (gCOD m− 3) MgCO3 Magnesite (kgCOD m− 3) MgHPO4 Newberyite Xli Lipids (ADM) (kgCOD m− 3) MgNH4PO4 Struvite XNOB Nitrite Oxidizing Bacteria (ASM) (gCOD m− 3) MLSS Mixed Liquor Suspended Solids XPP Polyphosphates (ASM, ADM) (gP m− 3) (kgP m− 3) MMP Multiple Mineral Precipitation Xpr Proteins (ADM) (kgCOD m− 3) N Nitrogen XS Biodegradable particulate organics (ASM) (gCOD m− 3) N2 Nitrogen gas XTSS Total Suspended Solids (ASM) (gTSS m− 3) N2O Nitrous oxide REC Recovery unit NaOH Sodium hydroxide 2 B. Solís et al. Water Research 215 (2022) 118223 modified Activated Sludge Model No. 2d (ASM2d) model with a speci­ 2. Material and methods ation model routine to predict pH at each time step (Flores-Alsina et al., 2015). This model evaluated and compared several energy and nutrient 2.1. BSM2-PSFe-GHG description recovery strategies, but without accounting for GHG emissions. Indeed, GHG emissions should be included when evaluating the 2.1.1. Biological models overall sustainability of control/operational strategies for water The ASM2d-PSFe-N2O model defined in this work merges the BSM2- resource recovery to add another important criterion in the multivari­ PSFe approach of Solon et al. (2017) and the ASM2d-N2O model of able space of performance assessment; otherwise, a good a priori control Massara et al. (2018) that has been successfully applied to describe N2O structure providing excellent effluent quality and lower costs could emissions in full-scale WWTPs (Solís et al., 2022). Hence, ASM2d-PS­ obtain this at the expense of high GHG emissions that are not being Fe-N2O describes simultaneous biological C, N and P removal, as well as considered. Previous modelling studies have already included GHG the chemical and biological processes related to S and Fe and N2O emissions as a potential performance criterion when evaluating the production and emission. Therefore, ASM2d-PSFe-N2O presents five sustainability of WWTPs. amongst the GHGs, N2O emissions, which have new state variables compared to the BSM2-PSFe model (i.e., SNO2, SNO, a 300-fold stronger global warming effect than carbon dioxide CO2 SN2O, SNH2OH and XNOB). The N2O biological pathways adapted from (IPCC, 2013), have recently received a lot of attention. Several exten­ Massara et al. (2018) are: sions based on ASM models have been proposed in the literature to better describe N2O emissions during biological nitrogen removal 1) NH2OH oxidation pathway (NN pathway): N2O is produced from the (Domingo-Félez et al., 2017; Mannina et al., 2016; Massara et al., 2018; reduction of NO by the enzyme “Nor” of AOB coupled with the Ni and Yuan, 2015; Pocquet et al., 2016). However, although some oxidation of NH2OH to NO2− (Pocquet et al., 2016); parameters of the models are pH-dependent, the evolution of pH in the 2) AOB nitrifier denitrification pathway (ND pathway): N2O is pro­ different reactors is not predicted since the effect on pH of the processes duced from NO2− reduction to NO and subsequently to N2O by AOB. taking place are not considered. Specifically the growth rate of nitrifiers These two processes are lumped in one single reaction as in Pocquet depends on pH, and consequently the N2O emissions produced by ni­ et al. (2016); trifiers cannot be described accurately for several operational conditions 3) heterotrophic denitrification pathway (DEN pathway): N2O is pro­ (Su et al., 2019). In addition, CO2 emissions are typically not accounted duced as an intermediate of the denitrification processes either by for, since the evolution of inorganic carbon (IC) is not modelled. How­ OHO or PAO (Hiatt and Grady, 2008). ever, nitrifiers growth depends on IC availability (Guisasola et al., 2007; Torà et al., 2010; Wett and Rauch, 2003; Zhang et al., 2018) and its The three biological N2O production pathways were included in the limitation could be significant in some scenarios. ASM2d-PSFe-N2O model to account for all the known biological N2O One of the most used plant-wide model that takes into account the production pathways and to fairly assess the contribution of each GHG emissions is the BSM2G (Flores-Alsina et al., 2011). Several works pathway under dynamic conditions and under the different control/ in the literature have applied this model to study the effect on GHG operational strategies implemented. The stoichiometric matrix and the emissions when implementing different control/operational strategies continuity verification of the modified ASM2d-PSFe-N2O model were (Barbu et al., 2017; Flores-Alsina et al., 2014, 2011; Santín et al., 2018, calculated as in Hauduc et al. (2010) and are provided in the Supple­ 2017; Sweetapple et al., 2015). However, BSM2G cannot describe the mentary Information Section. transformations and fate of P in the plant and, moreover, not all the The anaerobic digestion model (ADM) implemented is an extension known N2O production pathways are included in this model. Indeed, of the ADM1 model (Batstone et al., 2002), reproducing the biological N2O could be produced during the denitrifying phosphorus removal and chemical interactions between P, S and Fe as reported in previous process (Liu et al., 2015). Hence, a new model extension is needed to works (Flores-Alsina et al., 2016; Solon et al., 2017). The kinetic pa­ enable the evaluation of all the potential GHG emission sources when rameters of both models can be obtained from the software imple­ integrating the potential resource recovery mechanisms in WRRFs. mentation or from the original sub-models (Flores-Alsina et al., 2016; The current limitations of the previous approaches create the need to Massara et al., 2018; Solon et al., 2017). define a new extended benchmarking scenario (BSM2-PSFe-GHG) including biological COD/N/P removal, GHG emissions, and chemical 2.1.2. Physico-chemical models (PCMs) and physico-chemical models to evaluate resource recovery in WRRFs. BSM2-PSFe-GHG embraces three different PCMs as proposed in the The main objective of the present work is to develop and evaluate this BSM2-PSFe approach (Solon et al., 2017): the pH and ion speci­ BSM2-PSFe-GHG plant-wide benchmarking scenario by integrating: i) ation/pairing model (aqueous phase chemistry model), the multiple the biological model ASM2d-N2O proposed by Massara et al. (2018) mineral precipitation (MMP) model and the gas-liquid mass transfer accounting for both enhanced biological phosphorus removal (EBPR) model. and the most recently reported N2O production pathways, ii) potential sources of GHG emissions through the WRRF (updated from Flor­ 2.1.2.1. pH and ion speciation/pairing. A general aqueous phase chem­ es-Alsina et al. (2011)), iii) plant-wide modelling of detailed P chemical istry model is used in both ASM and ADM, describing the pH variation processes (Solon et al., 2017) and iv) development of novel control and ion pairing at each time step (Flores-Alsina et al., 2015; Solon et al., strategies based on nitrite and nitrous oxide sensors to mitigate N2O 2015). The aqueous phase chemistry model corrects for ionic strength emissions. Once the development of the BSM2-PSFe-GHG sub-models via the Davies’ approach for chemical activity (Solon et al., 2017). The and their interfaces is detailed, simulations will help to understand how acid-base parameters and the activity coefficients are novel nutrient recovery/control strategies can affect GHG emissions in a temperature-dependent and all calculations are performed under plant-wide context. In this sense, this work aims at i) studying the effect non-ideal conditions. The acid-base equilibria are described as a set of on GHG emissions when implementing nutrient recovery/control stra­ implicit algebraic equations and solved at each integration step of the tegies and ii) designing and implementing novel control/operational ordinary differential equation solver. The species concentrations take strategies to optimise plant performance while reducing the GHG part in the biological and physico-chemical processes. A more detailed emissions. description of the aqueous phase chemistry model can be found in Flores-Alsina et al. (2015) and Solon et al. (2015). The integration of the pH and ion speciation allows to account for weak acid-base conditions within the N2O production processes, since 3 B. Solís et al. Water Research 215 (2022) 118223 the growth rates of nitrifiers (XAOB and XNOB) are functions of their biodegradable particulate organics (XS) to Xpr, Xli and Xch; and (3) the substrates, i.e. free ammonia (FA, NH3) and free nitrous acid (FNA, direct mapping of acetate (SA to Sac) and inert soluble and particulate HNO2), respectively. organics (SI and XI) (Solon et al., 2017). Regarding the ADM→ASM interface, a comprehensive description of the involved processes and 2.1.2.2. Multiple mineral precipitation (MMP). The precipitation equa­ conversion can be found in Flores-Alsina et al. (2016). Finally, the PCMs tions are integrated as temperature dependent reversible processes with integration into ASM and ADM models was made following the pro­ the saturation index (SI) as the chemical driving force (Stumm and cedures detailed in the original works (Flores-Alsina et al., 2015; Solon Morgan, 1996). The SI was calculated by comparing the multiplication et al., 2017, 2015). of the chemical activities of the dissolved ions of each mineral with its solubility product. For a given aqueous phase, three conditions can 2.3. Plant layout and ancillary processes occur (Kazadi Mbamba et al., 2015a): i) SI < 0, the aqueous phase is undersaturated and the mineral is dissolved; ii) SI = 0, the aqueous BSM2-PSFe-GHG was implemented in the same plant layout as the phase is at equilibrium; or iii) SI > 0, the aqueous phase is oversaturated BSM2-PSFe (Solon et al., 2017). The WRRF consists of a primary clarifier and chemical precipitation may occur. The precipitation rate depends on (PRIM), an activated sludge section (AS), a secondary clarifier (SEC), a the kinetic rate coefficient, the species concentration, the mineral solid sludge thickener (THK), an anaerobic digester (AD), a dewatering unit phase and the order of the reaction (Kazadi Mbamba et al., 2015a, (DEW) and finally a storage tank (ST) (Fig. 1). Additional models were 2015b; Solon et al., 2017). The MMP model includes the most likely considered to simulate the ancillary processes PRIM, SEC, THK, DEW minerals to precipitate during wastewater treatment: calcite (CaCO3), and ST. The PRIM (900 m3) was modelled according to Otterpohl and hydroxyapatite (Ca5(PO4)3(OH)), amorphous calcium phosphate Freund (1992) with different settling velocities for biodegradable and (Ca3(PO4)2), struvite (MgNH4PO4⋅6H20), K-struvite (MgKPO4⋅6H20), non-biodegradable compounds (Wentzel et al., 2006). The AS had an newberyite (MgHPO4⋅3H20), magnesite (MgCO3) and iron sulphide A2/O configuration consisting of 7 tanks in series: Tanks 1 and 2 were (FeS). The simplified approach of Hauduc et al. (2015) is implemented anaerobic (ANAER1 and ANAER2) with a total volume of 2000 m3; tanks to describe the precipitation of hydrous ferric oxides (HFOs), the phos­ 3 and 4 were anoxic (ANOX1 and ANOX2) with a total volume of 3000 phate adsorption and phosphate co-precipitation to better estimate the m3 while tanks 5, 6 and 7 were aerobic (AER1, AER2 and AER3) with a phosphorus chemical precipitation. total volume of 9000 m3. The SEC (surface of 1500 m2 and height of 4 m) was modelled according to the double exponential settling velocity 2.1.2.3. Gas-liquid transfer. The gas-liquid transfer processes are function of Takács et al. (1991) in a ten-layer one-dimensional settler. described for the gas components: CO2, O2, NO, N2O, N2 and H2S. The The THK and DEW units were modelled as ideal units, with no biological gas-liquid transfer is based on Fick’s first law (Eq. (1)), which states that activity and a constant percentage of TSS in the concentrated sludge the transfer rate (ρi ) is proportional to the global mass transfer coeffi­ flows. The AD had a working volume of 3400 m3 and a headspace vol­ cient (kL ai ) and the driving force is the difference between the saturation ume of 300 m3. The ST was modelled as a non-reactive, ideally mixed concentration and the concentration of the gas in the liquid phase. The tank of 160 m3. Additional information about the plant design and saturation concentration is calculated through Henry’s law, which states default operational conditions can be found in Gernaey et al. (2014) and that there is a proportionality (KH,i) between the saturation concentra­ Solon et al. (2017). tion of the gas dissolved in the liquid and the partial pressure of the gas The influent was generated following the principles proposed by (Pi): Gernaey et al. (2011). Finally, the sensors and actuators were modelled ( ) with response time, delay and white noise to avoid creating unrealistic ρi = kL ai ⋅ KH,i ⋅Pi − Ci (1) control applications (Rieger et al., 2003). The mass transfer coefficient for each gas (i = CO2, O2, NO, N2O, N2 and H2S) is calculated from Eq. (2) as the square root of the ratio of the 2.4. Estimation of GHG emissions diffusivities of the gaseous component in the liquid (Di) to that of oxygen (DO2) and proportional to the mass transfer coefficient of the reference Different GHG compounds (CO2, CH4 and N2O) type of emissions compound oxygen (Lizarralde et al., 2015): (biogenic and non-biogenic) and sources of emissions (direct or indirect) were accounted for in BSM2-PSFe-GHG. Estimates not explicitly calcu­ ( )1/2 Di lated by the sub-models were estimated following the comprehensive kL ai = kL ao2 ⋅ (2) Do2 methodology suggested by Flores-Alsina et al. (2014, 2011). The different sources of GHG emissions considered throughout the WRRF The gas-liquid transfer processes in ADM are included for the are: following gas components: H2O, CO2, H2, CH4 and H2S, and are imple­ mented as described by Batstone et al. (2002). - Direct secondary treatment GHG emissions: CO2 generated from biomass respiration, CO2 generated from BOD5 oxidation, CO2 credit 2.2. Model integration from nitrification and N2O generated during biological N-removal. CO2 emissions are explicitly accounted for by ASM2d-PSFe-N2O and The different sub-models (ASM2d-PSFe-N2O, ADM and PCMs) in PCMs (i.e. pH and ion speciation/pairing and gas-liquid transfer BSM2-PSFe-GHG were integrated using model interfaces. The models), by including IC instead of alkalinity as a state variable ASM→ADM and ADM→ASM interfaces are based on the continuity- (Flores-Alsina et al., 2015). N2O generated via the NN and ND based interfacing method (Nopens et al., 2009) to ensure elemental pathways of AOB and DEN pathway of heterotrophic organisms mass and charge conservation. The interfaces consider instantaneous (Massara et al., 2018). processes (i.e. PROCESSAS-AD) and state variable conversions (i.e. - Sludge processing GHG emissions: GHG emissions during sludge pro­ CONVAS-AD). The ASM→ADM interface PROCESSAS-AD involves: (1) the cessing are generated in the anaerobic digester. CO2 and CH4 emis­ removal of COD demanding compounds (NH2OH, O2, NO3− , NO2− NO sions are explicitly calculated by the modified ADM1 model and N2O) with the associated growth of biomass, and (2) the decay of (Flores-Alsina et al., 2016; Solon et al., 2017). Fugitive emissions biomass (OHO, PAO, AOB and NOB) to produce proteins (Xpr), lipids from AD and co-generation units are included as a total of 2.7% of (Xli), carbohydrates (Xch) and inert particulate organics (XI). The CON­ the produced biogas that was slipped and un-combusted (Magnus VAS-AD involve (1) the conversion of soluble fermentable organics (SF) to Arnell, 2016). The remaining biogas is combusted in the gas-engine amino acids (Saa), sugars (Ssu) and fatty acids (Sfa); (2) the conversion of turbine and all the CH4 is converted to CO2, generating electricity 4 B. Solís et al. Water Research 215 (2022) 118223 Fig. 1. Layout of the WRRF from the BSM2-PSFe-GHG. GHG emissions (green box) and overall and individual mass balances (C, N and P) and pH for the main streams of the WRRF are indicated in the tables (steady state results for A0 open loop configuration). Inlet and outlet streams of the mass balances are highlighted in yellow and orange, respectively. and heat. The CO2 produced in the AD and the CO2 produced in the multiplying the truck movements by the distance of reuse. CO2 combustion are released into the atmosphere. Finally, dissolved CH4 emissions from mineralization are calculated based on the sludge (and H2) in the digester effluent is assumed to be fully stripped in the mass multiplied by the carbon concentration and the conversion following process units and emitted to the atmosphere. These emis­ factor from C to CO2. N2O emissions are calculated based on a static sions are accounted for in the AD (important to maintain mass bal­ factor of 0.01 kg N-N2O per kg of TN. In total, three different sludge ances). Potential N2O emissions due to denitrification downstream of disposal alternatives are included: Agriculture (38% sludge disposal, the AD unit (Oshita et al., 2014) were not considered since the DEW 150 km from the WRRF), Compost (45% sludge, 20 km) and Forestry unit was modelled as a non-reactive unit. Moreover, it should be (17% sludge, 144 km) (Arnell, 2016; Bridle et al., 2008; Flor­ noted that neither NO3− nor NO2− were present in the AD. es-Alsina et al., 2011). - Net power-related GHG emissions: Net power is the difference between energy consumption and production. Energy production is the elec­ Finally, all GHG emissions are converted into units of CO2 equiva­ tricity produced by the AD turbine and it is calculated using a factor lents (CO2e) by the Global Warming Potentials (GWP). The GWP for a for the energy content of CH4 (50 014 MJ (kg CH4)− 1) and assuming 100-year time horizon for N2O and CH4 are 298 kg CO2e per kg N2O and an efficiency of 43% for electricity generation (Flores-Alsina et al., 34 kg CO2e per kg CH4, respectively (IPCC, 2013). Additional details on 2011). Energy consumption involves pumping, mixing, aeration and the implementation of GHG emissions can be found in Section 6. heating and is calculated using the operational cost index (OCI), see Section 2.5. A value of 0.359 kg CO2 kWh− 1 is selected for the CO2 2.5. Evaluation criteria emission from net power production (European production mix) (IEA, 2011). Three performance indices were used to assess the plant performance - Embedded GHG emissions from chemicals use: The possible addition of for the different control/operational strategies. Besides the classical chemicals in the WRRF produces embedded indirect GHG emissions. evaluation criteria based on the effluent quality index (EQI) and the OCI The specific chemicals considered are: i) methanol dosage as external (Gernaey et al., 2014; Nopens et al., 2010), total GHG emissions (in carbon source with a static factor of 1.54 kg CO2 (kg methanol)− 1 CO2e) were added as an additional criterion, as first proposed by Flor­ (Flores-Alsina et al., 2011), ii) FeCl3 for P precipitation, 0.16 kg CO2 es-Alsina et al. (2014). This value enables the understanding of the (kg FeCl3)− 1, iii) NaOH to raise the pH, 1.24 kg CO2 (kg NaOH)− 1 and synergies and trade-offs that different nutrient recovery control strate­ iv) Mg(OH)2 to favour the struvite precipitation, 1.17 kg CO2 (kg Mg gies can have on overall GHG emissions. On the other hand, EQI (kg (OH)2)− 1 (Gustavsson and Tumlin, 2013). pollution units d− 1) represents the overall pollution leaving the plant - GHG emissions from effluent disposal: N2O is produced in the receiving and is calculated with Eq. (3) as a weighted sum of effluent TSS, COD, effluent due to the partial conversion of the remaining TN. An BOD, TKN, NOx (oxidized forms of nitrogen, including NO3− , NO2− , NO, emission factor of 5 g per kg TN discharged to recipient is obtained N2O and NH2OH) and organic and inorganic P (Porg and Pinorg, respec­ from the N2O emissions corresponding to disposal in lakes and rivers tively) (Solon et al., 2017). (Arnell, 2016). ∫ tstop - Sludge storage, disposal and reuse: Direct emissions from sludge stor­ EQI = 1 [β ⋅TSS(t) + βCOD ⋅COD(t) + βBOD ⋅BOD(t)+ age are estimated by assuming uncovered storage for 12 months as tobs 1000 tstart Tss ] (3) 8.68 kg CH4 per ton of VS and 1.1% of TN in sludge emitted as N2O βTKN ⋅TKN(t) + βNOx ⋅NOx (t) + βPorg ⋅Porg (t) + βPinorg ⋅Pinorg (t) ⋅Qe (t) ⋅dt (Arnell, 2016). After the sludge storage, it is transported for disposal and reuse, causing indirect emissions of CO2, CH4 and N2O. The CO2 where tobs is the total evaluation period, the βi are weighting factors for emissions associated with the transport of biosolids are quantified by the different pollutants to convert them into general pollution units 5 B. Solís et al. Water Research 215 (2022) 118223 (Solon et al., 2017) and Qe is the effluent flow rate in m3 d− 1. 3. Results The OCI is calculated with Eq. (4) as a weighted sum of the costs related to aeration, pumping, mixing and heating energy, external car­ 3.1. Steady-state simulations bon source, sludge production disposal, chemicals as well as the po­ tential benefits of methane production and nutrients recovered (e.g. Fig. 1 shows the total GHG emissions, combined with the fraction­ struvite). ation of GHG emissions (on-plant and off-plant), and the overall and individual mass balances for C, N and P as well as the pH under steady- OCI = AE + PE + f SP ⋅SP + f EC ⋅EC + ME − f MP ⋅MP + max(0, HE − 7MP)+ state conditions for the A0 scenario. amongst the total GHG emissions, f MA ⋅MA + f Mg ⋅Mg − f S recovered ⋅Srecovered 65% consisted of CO2 (of which 63% of the total CO2 emissions was (4) biogenic CO2 emitted in the biotreatment), 29% of N2O (21% of the total where AE is aeration energy, PE is pumping energy, SP is sludge pro­ N2O emitted was produced in the biotreatment section through N- duction, EC is external carbon addition, ME is mixing energy, MP is removal) and 6% of CH4. The low CH4 emissions were due to all the methane production, HE is required heating energy for the AD, MA is produced CH4 in the AD was burnt in the gas engine unit and, therefore, metal addition, Mg is magnesium addition and Srecovered is struvite transformed to CO2 and energy. Most of the GHG emissions were direct recovered. The fi are weighting factors and are selected as in Gernaey emissions (80%), i.e. produced in the WRRF. The predicted indirect GHG et al. (2014) and Solon et al. (2017). emissions were mainly produced due to sludge disposal and reuse, since Finally, other legal criteria, such as the percentage of time the plant the CO2 emissions produced due to electricity production were miti­ is in violation (TIV), i.e. when effluent concentrations are above gated from the electricity generated in the cogeneration unit of CH4 and discharge limits for selected nutrients in the effluent, were also used to no imbedded GHG emissions from chemicals use were produced. evaluate the plant performance. Regarding the fate of C, the inlet C ends up in three different forms: i) 51.3% is emitted as CO2: 32.8% in the AS section as biogenic CO2, due to 2.6. Control strategies the organic matter oxidation and biomass respiration and 18.5% as combustion and leakages of biogas in the AD (this represents 38.7% of Table 1 summarises the individual controllers and control strategies the inlet C to the digester), ii) 23.1% is dissolved in the effluent mainly in combining different controllers that were applied in this work. the form of SIC (80%) and SI (13.4%) and iii) 25.5% is disposed of in the Figures S1-S5 (Supplementary Information Section) show the sche­ sludge as particulate organics and biomass. matics of the control loops implemented in each control strategy. The In the case of N, the inlet N ends up in three different phases: i) 49% default scenario (A0) is the open-loop configuration (Gernaey et al., is discharged in the effluent mainly as SNO3 (31.4%) and dissolved SN2 2014), thus the air flow rate supplied to the aerobic reactors (value of (56.5%), ii) 21.4% ends up in the gas phase of the biological reactors, the mass transfer coefficient kLa) and the purge flow rate were kept mainly as N2, but with 1.0% of the inlet N as N–N2O, which is within the constant. The performance of each implemented control strategy is ranges reported by Massara et al. (2017) and Ahn et al. (2010) who evaluated by comparison with A0 by means of the evaluation criteria obtained values of 0–3.3% of N2O emission in 12 different WWTPs, and indices. The control strategies A1 to A3 are based on the improvement of iii) the remaining 29.6% of the inlet N is disposed in the sludge, mainly the water quality (reduction of EQI and TIV for N and P species) by as biomass and entrapped in particulate organics. One important optimizing the aeration strategy, the sludge age in winter or by outcome of this A0 operation is its feasibility to accomplish N-removal including nutrient recovery. Finally, the control strategies A4 and A5 are despite its lack of active control, since the values of TKN (2.8 g N m-3) mainly focused on reducing the GHG emissions while maintaining good and TN (11.0 g N m-3) in the effluent for A0 are below the BSM discharge effluent quality and low operating costs. limits (TKNlimit = 4 g N m− 3, TNlimit = 18 g N m− 3). The analysis of this All dynamic simulations (609 days) are preceded by a steady state scenario also shows the important effects of some recycled streams, such simulation (300 days) but only data generated during the last 364 days as the overflows of the thickener and the dewatering unit, which in­ of dynamic simulations are used to evaluate the implemented control crease the N influent load to the plant by 21.5%. strategies. The sensors characteristics applied in the implemented con­ Regarding the P results, only 22.3% of the influent P leaves the plant trol strategies are summarized in section S2. through the water line, mainly as soluble orthophosphate SPO4 (43.6%) and XPP (39.7%) that overflows in the secondary settler. The obtained effluent TP concentration is 2.37 g P m-3, above the BSM discharge limit of TPlimit = 2.0 g P m− 3. The remaining 77.7% of inlet P remain in the Table 1 Characteristics of the implemented controllers and control strategies. Controller→ DO NH4þ MLSS PO43¡ Magnesium Nitrite N2 O Characteristics↓ Measured variable SO2 in AER2 SNH4 in AER2 XTSS and T in AER3 SPO4 in Effluent SPO4 in REC SNO2 in AER2 SN2O in AER2 (s) AER3 unit Controlled variable SO2 in AER2 SNH4 in AER2 XTSS in AER3 SPO4 in XMg(OH)2 in REC unit SNO2 in AER2 SN2O in AER2 AER3 -3 -3 -3 - Set-point - gO m 2gNm 3000 g m (if T > 15 C) 4000 g m ◦ 1.0 g P m-3 50 g P m -3 0.5 g N m -3 0.01 g N m-3 3 (if T < 15 ◦ C) Manipulated kLa in AER1, 2 SO2 set-point in Qw QFeCl3 QMg(OH)2 SO2 set-point in SO2 set-point in variable &3 AER2 AER2 AER2 Control algorithm PI Cascaded PI PI PI PI Cascaded PI Cascaded PI Control strategy A0 A1 X X X A2 X X X X A3 X X X X A4 X X X X X A5 X X X X X 6 B. Solís et al. Water Research 215 (2022) 118223 waste sludge, pointing out the possibility of recovering P from the (SRT) and to maintain the nitrification capacity (Solon et al., 2017; anaerobic digestate. Moreover, the recycles of the thickener overflows Vanrolleghem et al., 2010). and the reject water, 7.7 and 229.9 kg P d− 1, respectively, increase the Table 2 shows that there is a reduction in N2O emissions due to the influent P load by 95%. increase of the DO-setpoint, which decreases the nitrite concentration compared to A0 and leads to a reduction of N2O emissions through the ND pathway Fig. 2.a shows that there are two different trends in N2O 3.2. Dynamic simulations emission rates depending on the season. On the one hand, the aeration demand is low during summer (day 254 to 357 and day 549 to 609), the The dynamic simulation results of the default A0 scenario and the DO ranges between 1 and 2 g O2 m− 3 and nitrite is accumulating in the runs with implemented control strategies A1-A5 are summarized in reactors (Fig. 2g). This causes N2O emissions via the ND pathway of Table 2. In the case of A0, 22.5% of the total GHG emissions come from AOBs to increase (Fig. 2d). On the other hand, during winter conditions, N2O during biological N-removal, which represents a N2O emission aeration increases and nitrite levels decrease, which deactivates the ND factor (N2O-EF) of 2.10%. This emission factor could be reduced by pathway. However, the production of N2O by the NN and DEN pathways analysing which biological pathways are producing most of the N2O increases because the cascade NH4+ control has difficulty in maintaining and, then, designing adequate mitigation strategies Table 2. also shows the desired NH4+ concentration during winter (see Figs. 2d and 2g) that the effluent obtained is acceptable in terms of effluent average considering the applied constraints in the DO set-point to avoid unre­ concentrations during the evaluated period. However, the percentages alistic control applications (minimum of 0 g O2 m-3 and maximum of 6 g of TIV for ammonium and P are high (35.3% and 40.5%, respectively) O2 m-3). The GHG emissions from the biotreatment (CO2 biogenic plus and thus, there is a niche for a performance improvement using control N2O from N-removal) and the total GHG emissions decreased (4.0% and strategies. In the following sections, the results for each implemented 3.6%, respectively), due to the decrease in N2O emissions. The variation control strategy (Table 1) are presented and discussed. The GHG emis­ of the waste flow rates during summer and winter led to an improvement sions and the overall and individual mass balances (Q, C, N and P) and in the AD performance, since more methane was produced (Eproduction pH for the main streams of the WRRF for each control strategy are re­ increased), which however led to an increase in AD emissions due to ported in sections S3 and S4 (Supplementary Information Section). increased combustion of biogas. EQI improved in A1 due to lower effluent N concentrations: TKN 3.2.1. Control strategy A1: Ammonium cascade & waste controller decreased from 5.8 to 3.6 g N m− 3 (A0 vs A1) and the TIV of ammonium The A1 control strategy involves three controllers. The first two decreased from 35.3 to 0.2%. The average P concentration remained the control loops include two controllers following a cascade configuration, same and the total P concentration in the effluent decreased by only 0.1 currently known as aeration-based ammonium controller (ABAC). In g P m-3 compared to A0. The OCI increased compared to A0 mainly due to this configuration, the DO controller of the secondary feedback control increased aeration costs during the winter period (i.e. when the tem­ loop is in charge of maintaining the DO concentration in AER2 by perature is below 15 ◦ C, between days 357 and 549 of the simulation), manipulating the aeration flow (kLa value), while the primary feedback since a higher DO set-point is required to maintain the desired ammo­ control loop manipulates the DO set-point in AER2 using the ammonium nium concentration (Fig. 2g). concentration in AER2 as the controlled variable. The ammonium set- point in AER2 reactor is fixed at 2 g N m− 3. An additional control loop 3.2.2. Control strategy A2: Fe chemical precipitation of PO43- acts on the purge flow (Qw) to maintain the desired XTSS concentration in Control strategy A2 aims at reducing the effluent P concentration via AER3. The XTSS set-point depends on the temperature (Table 1). The XTSS its chemical precipitation with Fe by adsorption and co-precipitation of concentration is increased from 3 000 to 4 000 g TSS m-3 during winter phosphate species onto HFOs. A2 includes A1 and a PI controller that conditions (i.e. T < 15 ◦ C) to establish a longer sludge retention time Table 2 Performance evaluation criteria for each control strategy. Control strategy → A0 A1 A2 A3 A4 A5 units Emitted CO2 biogenic 7 467 7 510 7 616 7 470 7 569 7 527 kg CO2e d− 1 Emitted N2O N-removal 5 237 4 681 4 685 4 312 3 987 3 832 kg CO2e d− 1 N2O-EF total 2.10 1.33 1.35 1.27 1.17 1.11 % Total emissions biotreatment 12 703 12 191 12 301 11 782 11 556 11 359 kg CO2e d− 1 AD emissions 4 366 4 462 4 528 4 252 4 238 4 261 kg CO2e d− 1 Total GHG emissions 23 339 22 494 22 844 22 363 21 333 21 164 kg CO2e d¡1 Direct GHG emissions 18 970 18 582 18 796 17 743 17 491 17 326 kg CO2e d− 1 Indirect GHG emissions 4 369 3 912 4 049 4 620 3 842 3 837 kg CO2e d− 1 Nkjeldahl 5.8 3.6 3.5 3.8 3.6 3.6 g N m− 3 Ntotal 13.0 11.3 11.4 10.6 10.9 10.9 g N m− 3 Pinorg 1.0 1.0 0.5 0.1 0.1 0.1 g P m− 3 Ptotal 2.5 2.4 1.8 0.9 0.9 0.9 g P m− 3 TIV SNH4 (= 4 g N m− 3) 35.3 0.2 0.6 0.2 0.1 0.1 % TIV Ntotal (= 18 g N m− 3) 0.2 0.0 0.0 0.0 0.0 0.0 % TIV Ptotal (= 2 g P m− 3) 40.5 34.1 20.0 0.3 0.3 0.3 % EQI 11 769 10 338 9 074 7 129 7 240 7 238 kg p.u. d¡1 Eaeration 4 000 4 445 4 838 4 031 4 126 4 237 kWh d− 1 Eproduction a 5 674 5 791 5 897 5 906 5 829 5 860 kWh d− 1 SPdisposal 4 033 4 068 4 532 3 643 3 632 3 641 kg TSS d− 1 QFeCl3 b 0 0 88 0 0 0 kg Fe d− 1 QMg(OH)2 b 0 0 0 80 80 80 kg Mg d− 1 Srecovered c 0 0 0 442 442 442 kg struv d-1 OCI 11 864 12 306 16 109 10 045 10 224 10 362 – a Energy production. The electricity generated by the turbine, calculated as the energy content of methane gas. b Relative costs for FeCl3, Mg(OH)2 and recovered struvite are the same as in Solon et al. (2017). c Srecovered refers to recovered struvite. 7 B. Solís et al. Water Research 215 (2022) 118223 Fig. 2. Dynamic profiles of control strategies A1 (a, d, g), A4 (b, e, h) and A5 (c, f, i). (a, b, c): N2O emissions in the AS unit; (d, e, f): N2O production rates in AER2 reactor and (g, h, i): nitrite, ammonium and DO con­ centrations in AER2 reactor. A 3-day first-order expo­ nential filter is used to improve visualization of the results. Negative values of the DEN pathway mean that, for the 4-step denitrification process, the rate of N2O reduction to N2 is higher than that of N2O production from NO. Additional dynamic profiles for all the control strategies are shown in section S4 of SI. regulates the FeCl3 addition in AER3 reactor to maintain the P concen­ of Fe, compared to control strategy A0 (Table 2). However, the controller tration in AER3 reactor at the desired set-point of 1 g P m− 3 (Table 1). was not able to maintain the SPO4 at the desired set-point. The average The average SPO4 concentration in scenario A0 already was 1 g P⋅m− 3, FeCl3 flow rate throughout the evaluation period was 88 kg Fe/d, which but high P peaks were observed in the effluent. The objective of A2 is led to a considerable increase of the operational cost, mainly due to the mitigating these P peaks avoiding the high TIV = 40.5% observed. iron dosage (2400 $ (Ton Fe)− 1, (Solon et al., 2017)). Regarding GHG emissions, total emissions in A2 increased slightly due to i) more biogenic CO2 was emitted: PAO activity decreased 3.2.3. Control strategy A3: Struvite recovery because there was less phosphate in the anaerobic reactor, resulting in a Control strategy A3 complements A1 by including P- (and N-) re­ higher fraction of COD removed by heterotrophic biomass; this biomass covery as struvite in the digester supernatant. The layout of the WRRF produces more inorganic carbon than PAO when removing COD, ii) was modified by including a recovery unit (REC) based on struvite higher production of biogas and therefore higher emissions from the AD: precipitation (see Figure S3 in Supplementary Information Section). The the iron species enhance primary clarification and more COD is redir­ REC unit includes a crystallizer to support struvite precipitation, a ected to the AD system and iii) indirect CO2 emitted by the use of FeCl3. storage tank for magnesium hydroxide (Mg(OH)2) and a dewatering unit The observed N2O emissions were the same as in control strategy A1 (Kazadi Mbamba et al., 2016; Solon et al., 2017). A PI controller was because the N fluxes were not affected by the addition of iron (see added to control the effluent P from the recovery unit at a set-point of 50 Table 2). g P m-3 by manipulating the Mg(OH)2 flow rate (QMg(OH)2). A2 led to a lower concentration of P in the effluent and, consequently, GHG emissions from the whole WRRF decreased (Table 2). N2O the TIV of total P decreased from 40.5% with A0 to 20.0% with A2 and emissions decreased slightly because the influent N load to the AS unit the EQI was reduced by about 23% (Table 2). The phosphate controller decreased due to struvite crystallization in the reject water stream and, was able to reduce the SPO4 peaks in the AER3 reactor with the addition thus, P- and N-recovery as struvite also had a potential benefit on GHG 8 B. Solís et al. Water Research 215 (2022) 118223 emissions due to more diluted streams. The struvite recovered was 442 cascade PI controller. Both controllers calculated an adequate DO set- kg d− 1, which resulted in 99.8 kg P d− 1 (48.5% of the total P influent point and the maximum value was chosen (see Table 1 for the charac­ load) and 45.0 kg N d− 1 recovered (4.2% of the total N influent load), teristics of the controllers). The set-point signal of both controllers was respectively. The reject water P load was reduced from 232.3 kg P d− 1 smoothened using a first-order exponential filter with a time constant of (A1) to 11.2 kg P d− 1 (A3), which resulted in a 95% reduction in the 15 min to avoid numerical instabilities during solver integration. influent P load to the biological reactors. A4 led to the minimum GHG emissions with respect to the previously Table 2 shows that the average effluent P concentrations in A3 were implemented strategies (A0 to A3): the N2O emissions were reduced by lower than those in strategies A0 to A2: the WRRF was able to discharge P 7.5% compared to A3. The implementation of the nitrite PI cascade below the legal limits most of the time (TIV of 0.3%) and EQI decreased controller reduced the N2O emissions during the summer conditions by a significant 31% with respect to A1. Table 2 also shows that OCI compared to A1 (Figs. 2a and 2b), since one of the substrates of the ND decreased 18% compared to A1, i.e. struvite recovery is techno- pathway, i.e. nitrite, was minimized (Figs. 2d and 2e). The N2O emis­ economically feasible considering only the operational costs associ­ sions during winter conditions remained the same as in A1 because the ated with the addition of Mg and struvite revenues in a current market ammonium PI cascade was preferentially fixing the DO set-point. The scenario. More struvite could be recovered by lowering the phosphate nitrification capacity should be increased in order to further reduce the set-point of the controller, since there was still a surplus of 11.2 kg d− 1 of N2O emissions during winter by, for example, increasing the DO levels inorganic P available to be precipitated as struvite (Fig. 3a). However, or the MLSS concentration, with the trade-off of further increasing the this would imply a higher cost of Mg(OH)2 and with the selected setpoint operational costs. it was enough to meet P discharge limits. Struvite can be precipitated in A4 slightly increased the effluent N concentration in comparison to a wide range of pH (between 7 and 11) with an optimum pH range A3 (2.8% increase in total N compared to A3) due to more ammonium between 8.0 and 9.5. The addition of Mg(OH)2 was enough to increase being nitrified in A4 compared to A3 (effluent TKN decreased by 5%) and pH from 7.1 to 8.3 (Fig. 3b) favouring struvite precipitation without the increase of effluent nitrate concentration. In this sense, the imple­ requiring an additional aeration unit for CO2 stripping nor the addition mentation of ammonium and nitrite cascade controllers also slightly of more alkalinity such as NaOH (Kazadi Mbamba et al., 2016; Solon increased OCI by 1.8%, compared to A3, since the applied DO set-point et al., 2017). Further studies are required to assess the capital costs was always the maximum of the ammonium and nitrite controllers and associated with struvite recovery and additional transport costs (these the aeration costs incremented by 2.3% compared to A3. The same costs were not considered in the evaluation criteria). amount of struvite was obtained as in A3 because the fluxes of P in the sludge line remained unaffected. Fig. 2h shows that during summer 3.2.4. Control strategy A4: Ammonium & nitrite cascade controllers and conditions (i.e. T above 15 ◦ C) the DO set-point is mostly defined by the struvite recovery nitrite controller (NH4+ is below the set-point of 2 g N m− 3 and NO2− Control strategy A4 aims at reducing GHG emissions with a particular concentration is around the set-point of 0.5 g N m− 3). The NH4+ emphasis on N2O emissions derived from biological N-removal. A4 ex­ controller is only activated during the daily peaks when the influent N tends A3 with a cascade PI nitrite controller in AER2 reactor. Nitrite load is high (in summer the DO set-point is defined by the NH4+ concentration was maintained at the desired set-point by manipulating controller only 23% of the time). On the other hand, during winter the set-point of the DO controller in conjunction with the ammonium conditions the DO set-point is defined most of the time (62%) by the NH4+ controller to ensure complete nitrification. 3.2.5. Control strategy A5: Ammonium & nitrous oxide cascade controllers and struvite recovery A5 is a modification of A4 that also aimed at reducing N2O emissions. New sensors have appeared in the market that enable the monitoring of soluble N2O concentration in the reactors with high accuracy and, thus, allow designing novel mitigation strategies. For this reason, A5 included a cascade PI controller based on the measurement and control of N2O concentration in AER2. In a similar way to A4, N2O and NH4+ controllers calculated SO2 set-points for the DO controller and the chosen value was the maximum (Table 1). The GHG emissions obtained were the lowest amongst all the control Fig. 3. a) Total P effluent load of the recovery unit (returns to water line) for control strategies A1 and A3. b) Simulated pH values of the recovery unit influent prior and post magnesium addition for control strategy A3. A 3-day first-order exponential filter is used to improve the visualization of the re­ Fig. 4. Simulated soluble N2O concentration in AER2 for A4 and A5. A 3-day sults. Raw data is shown in grey. first-order exponential filter is used to improve the visualization of the results. 9 B. Solís et al. Water Research 215 (2022) 118223 strategies implemented (Table 2), with 13% reduction in N2O emissions Control strategies A4 and A5 merged the ammonium cascade controller compared to A1, and 1% reduction compared to A4. Fig. 4 shows the of A1 with another nitrite or soluble N2O cascade controller and the concentration of soluble N2O in AER2 predicted in A4 and A5. N2O struvite precipitation of A3. Both control strategies led to higher oper­ concentration in A5 was much more constant due to the N2O PI cascade ational costs than A3, 1.8 and 2.8%, respectively, due to the increased controller, which actively imposed the DO set-point when the N2O aeration demand imposed by the cascade controllers. A5 seems to have a concentration in AER2 was too high. Only in the transition period from better performance since it led to a reduction of the emitted N2O in the summer to winter (T around 15 ◦ C), the cascade PI controllers of A4 biotreatment of 27% but at the expense of higher costs (i.e. 1.3% higher achieved lower N2O concentrations (and lower N2O emissions) than in in A5 compared to A4). There is therefore a compromise between oper­ A5. This was due to the N2O PI of A5 being deactivated and the DO ational costs and GHG emissions, since operational costs increased setpoint was being fixed by the NH4+ controller. On the other hand, slightly in both strategies compared to A3, where the main difference during summer and winter, i.e. T above ~16 ◦ C and T below ~14 ◦ C, the between the objectives of A3 compared to A4 and A5 was the reduction of cascade controllers of A5 achieved lower N2O soluble concentration in GHG emissions, and moreover, A4 and A5 achieved the same EQI. AER2 (Fig. 4) and lower global N2O emissions (Fig. 2b and c). These Finally, Fig. 5 shows that the largest reduction in total GHG emis­ results also point to a simpler alternative strategy that could lead to sions was 9% compared to A0, despite the fact that the main aim of the reduction in N2O and GHG emissions. Reducing the NH4+ set-point in novel control strategies is N2O reduction. Other sources of GHG emis­ strategy A1 or implementing a properly selected temperature-dependent sions were not reduced, such as indirect emissions (electricity, chemical NH4+ set-point (similar to the TSS controller) would probably provide usage, sludge storage and reuse) which represented about 20% of the some further improvements, although the automated set-point selection total GHG emissions, and other direct GHG sources that were not of the closed-loop control would be lost in this case. controllable, such as biogenic CO2 and methane combustion, which The effluent concentrations and the TIV obtained for A5 were the together represented around 50% of the total GHG emissions (see same as for A4, therefore, a high effluent quality was obtained. However, Table S1 of Supplementary Information Section). the OCI was 1.3% higher than A4, since slightly more oxygen was required to maintain the N2O set-point: the DO set-point was set by the 4. Comparison with other works and limitations of the proposed N2O cascade PI in A5 58% of the time whereas the DO set-point was fixed methodology by the nitrite controller in A4 56% of the time. The proposed BSM2-PSFe-GHG plant-wide model and the imple­ 3.2.6. Comparison of the evaluation criteria for the control strategies mented control strategies results represent an improvement to the cur­ implemented rent BSM modelling framework BSM2-PSFe (Solon et al., 2017) by Fig. 5 compares EQI, OCI, biogenic N2O emissions and total GHG adding the GHG production and emission during nutrient removal and emissions for each control strategy implemented. The data are normal­ recovery operational/control strategies. In this sense, the ised considering 100% for the values obtained with the reference BSM2-PSFe-GHG provides a new tool that shows, in a plant-wide operation A0. All control strategies led to a more sustainable overall context, the trade-offs that different novel control strategies had on plant performance, since all of them obtained a better effluent quality (i. the sustainability of the WRRF. On the other hand, the BSM2-PSFe-GHG e. lower EQI) and lower GHG emissions compared to the default sce­ updates previous works addressed to characterize GHG emissions, with a nario. Regarding operational costs, the ammonium cascade controller particular emphasis on N2O emissions, which were designed for (A1) increased the OCI by 4% compared to A0 due to the more intense different plant-wide models (Flores-Alsina et al., 2014, 2011; Sweet­ aeration demands. The chemical P precipitation strategy (A2) increased apple et al., 2014) by: i) adding the GHG emissions to the most recent the OCI by 36% compared to A0 due to the high cost of FeCl3 dosage. On BSM modelling framework capable of simulating nutrient recovery the other hand, struvite precipitation in the reject water (control strat­ strategies (Solon et al., 2017), ii) adding all the known biological N2O egy A3) was the most successful strategy in terms of EQI and OCI, leading pathways reported in the ASM2d-N2O model (Massara et al., 2018), iii) to a reduction in EQI of 40% compared to A0 and 31% compared to A1, improving the calculation of CO2 emissions by including the general and a reduction in OCI of 11% and 14% compared to A0 and A1, aqueous phase model (Flores-Alsina et al., 2015; Solon et al., 2017) and respectively. These improvements were due to: 1) the potential benefits iv) updating the sources of direct and indirect GHG emissions (Flor­ of struvite sales and 2) the reduction in influent load of P and N, which es-Alsina et al., 2011; Arnell, 2016). led to lower aeration demand. Control strategies A4 and A5 obtained The results reported for each control strategy were unified into three higher reduction in N2O emission from N-removal compared to A0. main groups (EQI, OCI and GHG) as proposed by Flores-Alsina et al. Fig. 5. Comparison of the evaluation criteria for the control strategies implemented. Data is shown in relative percentage compared to control strategy A0. 10 B. Solís et al. Water Research 215 (2022) 118223 (2014). This enabled a fairer evaluation of the different strategies, since would increase the capital and maintenance costs of the associ­ none of the criteria depended on the others. Other works have proposed ated sensors, instruments and controllers. to unify the multicriteria into a single cost function by transforming effluent quality into monetary units by defining tariffs or taxes when the 5. Conclusions concentrations in the effluent are above a certain limit (Guerrero et al., 2012; Stare et al., 2007), or applying a defined weighted average of the In this paper, a novel plant-wide model that integrates the latest different evaluation criteria (Machado et al., 2020). The benefits of the advances in energy and nutrient recovery modelling for an accurate latter approach are that control strategies are compared with a single description of N2O- and EBPR-related processes is proposed. Five control index. In this work, a unified cost function could be defined if GHG strategies are evaluated in view of optimising plant performance, emissions were also translated into monetary units, by imposing tariffs minimizing GHG emissions and implementing nutrient recovery. The due to high emissions. Special attention should be paid to defining the main findings of the work are: different weights of the cost function, since an optimisation of the cost function could lead to high GHG emissions or to poor effluent quality - Direct and indirect GHG emissions for CO2, N2O and CH4 were with low operational costs. quantified in the whole WRRF. P recovery as struvite (strategies A3 onwards) showed an improve­ - Overall and individual mass balances quantify the distribution of C, ment in the operational costs due to the potential revenues from struvite; N and P in the whole WRRF. however, these results should be taken with caution as the assumed price - All five control strategies led to an overall more efficient and sus­ of struvite (200 $ ton− 1 as in Solon et al. (2017)) is very uncertain. In tainable plant performance. addition, struvite recovery improved effluent quality due to reduced P - P-recovery as struvite led to decreased P and N concentrations in the and N loading in the sludge line recycles and that also decreases oper­ biological reactors which reduced the N2O emissions in the bio­ ational costs. In fact, the OCI would also improve by 17.6%, compared to treatment by 17%, compared to the open loop configuration. A1, assuming no benefit from struvite sales. Finally, P recovery as stru­ - The lowest N2O and overall GHG emissions were achieved when vite also showed a reduction in GHG emissions from the AS unit, mainly ammonium and soluble nitrous oxide in the aerobic reactors were due to the decrease in the N influent load. Other important assumptions controlled, achieving a reduction of 24% and 27% for N2O, respec­ were made in the crystallizer unit model, such as ideal solids separation tively, and 9% for total GHG, compared to the open loop and simplified precipitate dissolution (Solon et al., 2017). In addition, configuration. potential pipe clogging in the REC unit due to struvite precipitation was not considered and is known to be a major issue during P recovery as Software availability struvite. These limitations in the crystallizer model should be addressed in future work to obtain a better estimation of struvite recovery. The MATLAB/SIMULINK code of the models presented in this work is One limitation of BSM2-PSFe-GHG is that capital expenditure was available upon request. Using this code, readers will be able to repro­ not included in the evaluation criteria and the comparison between duce the results included in this paper. To express interest, please con­ control strategies was only subject to operational costs. Adding the tact with Dr. Juan Antonio Baeza (JuanAntonio.Baeza@uab.cat) or Dr. capital costs of equipment, sensors, civil, electrical and piping will Albert Guisasola (Albert.Guisasola@uab.cat). provide a more complete assessment (Machado et al., 2020; Ostace et al., 2013; Solon et al., 2017). For example, integrating P-recovery as Declaration of competing interest struvite recovery implies a modification of the plant layout or adding a REC unit and all related equipment. That would result in a higher capital The authors declare that they have no known competing financial investment when retrofitting or upgrading the WRRF. On the other interests or personal relationships that could have appeared to influence hand, P precipitation by Fe addition (control strategy A2), showed the work reported in this paper. higher operational costs than A3 but this strategy “only” implies adding an extra dosing tank to the existing plant layout. Acknowledgements The proposed control strategies showed the logical steps that a WRRF manager should take to improve effluent quality (A1 to A3) and, after­ The authors are very grateful to Dr Kimberly Solon (Ghent Univer­ wards, to reduce GHG emissions (A4 and A5). However, each of the sity, Belgium) for providing the BSM2-PSFe software code. Borja Solís is control strategies could be optimized: grateful for the PIF PhD grant funded by Universitat Autònoma de Barcelona. Borja Solís, Albert Guisasola and Juan Antonio Baeza are i) the location of the Fe addition in A2 can be optimised to reduce members of the GENOCOV research group (Grup de Recerca Consolidat operational costs as already reported (Kazadi Mbamba et al., de la Generalitat de Catalunya, 2017 SGR 1175, www.genocov.com). 2019); Xavier Flores-Alsina gratefully acknowledges the financial support ii) each of the set-point values can be optimized as in Guerrero et al. provided by the Danish International Development Agency (Danida (2011) in order to decrease the EQI and OCI, and to minimise Fellowship center) in the frame of the ERASE project (grant number 18- GHG emissions. For instance, the reduction of the NH4+ set-point M09-DTU) and the Danish MUDP (Miljøteknologisk udviklings- og value from 2.0 to 1.0 g N m− 3 in A1 (Table 1) led to a 45% demonstrationsprogram) under the project NACAT. A partial pre­ reduction in N2O emissions, while the OCI increased by 4% and liminary version of this paper was presented at Watermatex 2019 the EQI by 14%. (Copenhagen, Denmark, September 2019). iii) N2O emissions in the AS unit could be reduced by adding DO, NH4+, NO2− or N2O sensors and controllers in each aerobic Supplementary materials reactor to better control the WRRF as in Santín et al. (2017), who also aimed at reducing GHG emissions during wastewater treat­ Supplementary material associated with this article can be found, in ment by using the BSM2G modelling framework (Flores-Alsina the online version, at doi:10.1016/j.watres.2022.118223. et al., 2011). This strategy enabled a more robust DO control and, therefore, a more robust control of N2O emissions. However, the addition of multiple controllers in each aerobic reactor results in a more complex control structure for the biological reactors and 11 B. Solís et al. Water Research 215 (2022) 118223 References Hiatt, W.C., Grady, C.P.L., 2008. An Updated Process Model for Carbon Oxidation, Nitrification, and Denitrification. Water Environ. Res. 80, 2145–2156. https://doi. org/10.2175/106143008X304776. Ahn, J.H., Kim, S., Park, H., Rahm, B., Pagilla, K., 2010. N2O Emissions from activated IEA, 2011. Emissions from Fuel combuistion. Highlights. Technical report. International sludge processes, 2008 - 2009 : results of a national monitoring survey in the United Energy Agency, Paris, France n.d. States. Environ. Sci. Technol. 44, 4505–4511. https://doi.org/10.1021/es903845y. IPCC, 2013. Climate change 2013: the physical science basis. In: Contribution of Working Amaral, A., Gillot, S., Garrido-Baserba, M., Filali, A., Karpinska, A.M., Plósz, B.G., De Group I to the Fifth Assessment Report of the Intergovern - mental Panel on Climate Groot, C., Bellandi, G., Nopens, I., Takács, I., Lizarralde, I., Jimenez, J.A., Fiat, J., Change. WGI [Stocker, T.F., D. Qin, G.-K. Plattner, M. Tignor, S.K. Allen, J. Rieger, L., Arnell, M., Andersen, M., Jeppsson, U., Rehman, U., Fayolle, Y., Boschung, A., n.d. Amerlinck, Y., Rosso, D., 2019. Modelling gas–liquid mass transfer in wastewater Jeppsson, U., Alex, J., Batstone, D.J., Benedetti, L., Comas, J., Copp, J.B., Corominas, L., treatment: when current knowledge needs to encounter engineering practice and Flores-Alsina, X., Gernaey, K.V., Nopens, I., Pons, M.-.N., Rodríguez-Roda, I., vice versa. Water Sci. Technol. 80, 607–619. https://doi.org/10.2166/ Rosen, C., Steyer, J.-.P., Vanrolleghem, P.A., Volcke, E.I.P., Vrecko, D., 2013. wst.2019.253. Benchmark simulation models, quo vadis? Water Sci. Technol. 68, 1–15. https://doi. Arnell, M., 2016. Performance Assessment of Wastewater Treatment Plants. Lund org/10.2166/wst.2013.246. University. Kazadi Mbamba, C., Batstone, D.J., Flores-Alsina, X., Tait, S., 2015a. A generalised Barat, R., Serralta, J., Ruano, M.V., Jiménez, E., Ribes, J., Seco, A., Ferrer, J., 2013. chemical precipitation modelling approach in wastewater treatment applied to Biological nutrient removal model no. 2 (BNRM2): a general model for wastewater calcite. Water Res. 68, 342–353. https://doi.org/10.1016/j.watres.2014.10.011. treatment plants. Water Sci. Technol. 67, 1481–1489. https://doi.org/10.2166/ Kazadi Mbamba, C., Flores-Alsina, X., John Batstone, D., Tait, S., 2016. Validation of a wst.2013.004. plant-wide phosphorus modelling approach with minerals precipitation in a full- Barbu, M., Vilanova, R., Meneses, M., Santin, I., 2017. On the evaluation of the global scale WWTP. Water Res. 100, 169–183. https://doi.org/10.1016/j. impact of control strategies applied to wastewater treatment plants. J. Clean. Prod. watres.2016.05.003. 149, 396–405. https://doi.org/10.1016/j.jclepro.2017.02.018. Kazadi Mbamba, C., Lindblom, E., Flores-Alsina, X., Tait, S., Anderson, S., Saagi, R., Batstone, D.J., Keller, J., Angelidaki, I., Kalyuzhnyi, S.V., Pavlostathis, S.G., Rozzi, A., Batstone, D.J., Gernaey, K.V., Jeppsson, U., 2019. Plant-wide model-based analysis Sanders, W.T.M., Siegrist, H., Vavilin, V.A., 2002. Anaerobic Digestion Model No. 1. of iron dosage strategies for chemical phosphorus removal in wastewater treatment IWA Sci. Tech. Rep. n.13. IWA Publ, London, UK. systems. Water Res. 155, 12–25. https://doi.org/10.1016/j.watres.2019.01.048. Bridle, T., Shaw, A., Cooper, S., Yap, K.C., Third, K., 2008. Estimation of greenhouse gas Kazadi Mbamba, C., Tait, S., Flores-Alsina, X., Batstone, D.J., 2015b. A systematic study emissions from wastewater treatment plants. In: Proceedings IWA Wordl Water of multiple minerals precipitation modelling in wastewater treatment. Water Res. Congress 2008, Vienna, Austria. September 7-12, 2008. 85, 359–370. https://doi.org/10.1016/j.watres.2015.08.041. Domingo-Félez, C., Calderó-Pascual, M., Plósz, B.G., Smets, B.F., Sin, G., Plósz, B.G., Liu, Y., Peng, L., Chen, X., Ni, B.J., 2015. Mathematical Modeling of Nitrous Oxide 2017. Calibration of the comprehensive NDHA-N2O dynamics model for nitrifier- Production during Denitrifying Phosphorus Removal Process. Environ. Sci. Technol. enriched biomass using targeted respirometric assays. Water Res. 126, 29–39. 49, 8595–8601. https://doi.org/10.1021/acs.est.5b01650. https://doi.org/10.1016/j.watres.2017.09.013. Lizarralde, I., Fernández-Arévalo, T., Brouckaert, C., Vanrolleghem, P., Ikumi, D.S., Flores-Alsina, X., Arnell, M., Amerlinck, Y., Corominas, L., Gernaey, K.V., Guo, L., Ekama, G.A., Ayesa, E., Grau, P., 2015. A new general methodology for Lindblom, E., Nopens, I., Porro, J., Shaw, A., Snip, L., Vanrolleghem, P.A., incorporating physico-chemical transformations into multi-phase wastewater Jeppsson, U., 2014. Balancing effluent quality, economic cost and greenhouse gas treatment process models. Water Res. 74, 239–256. https://doi.org/10.1016/j. emissions during the evaluation of (plant-wide) control/operational strategies in watres.2015.01.031. WWTPs. Sci. Total Environ. 466–467, 616–624. https://doi.org/10.1016/j. Machado, V.C., Lafuente, J., Baeza, J.A., 2020. Systematic comparison framework for scitotenv.2013.07.046. selecting the best retrofitting alternative for an existing water resource recovery Flores-Alsina, X., Corominas, L., Snip, L., Vanrolleghem, P.a, 2011. Including greenhouse facility. Water Environ. Res. 92, 2072–2085. https://doi.org/10.1002/wer.1368. gas emissions during benchmarking of wastewater treatment plant control strategies. Magnus Arnell, 2016. Performance Assessment of Wastewater Treatment Plants. Multi- Water Res. 45, 4700–4710. https://doi.org/10.1016/j.watres.2011.04.040. Objective Analysis Using Plant-Wide Models. Flores-Alsina, X., Kazadi Mbamba, C., Solon, K., Vrecko, D., Tait, S., Batstone, D.J., Mannina, G., Ekama, G., Caniani, D., Cosenza, A., Esposito, G., Gori, R., Garrido- Jeppsson, U., Gernaey, K.V., 2015. A plant-wide aqueous phase chemistry module Baserba, M., Rosso, D., Olsson, G., 2016. Greenhouse gases from wastewater describing pH variations and ion speciation/pairing in wastewater treatment process treatment - a review of modelling tools. Sci. Total Environ. 551–552, 254–270. models. Water Res. 85, 255–265. https://doi.org/10.1016/j.watres.2015.07.014. https://doi.org/10.1016/j.scitotenv.2016.01.163. Flores-Alsina, X., Solon, K., Kazadi Mbamba, C., Tait, S., Gernaey, K.V., Jeppsson, U., Massara, T.M., Malamis, S., Guisasola, A., Baeza, J.A., Noutsopoulos, C., Katsou, E., Batstone, D.J., 2016. Modelling phosphorus (P), sulfur (S) and iron (Fe) interactions 2017. A review on nitrous oxide (N2O) emissions during biological nutrient removal for dynamic simulations of anaerobic digestion processes. Water Res. 95, 370–382. from municipal wastewater and sludge reject water. Sci. Total Environ. 596–597. https://doi.org/10.1016/j.watres.2016.03.012. https://doi.org/10.1016/j.scitotenv.2017.03.191. Gernaey, K.V., Flores-Alsina, X., Rosen, C., Benedetti, L., Jeppsson, U., 2011. Dynamic Massara, T.M., Solís, B., Guisasola, A., Katsou, E., Baeza, J.A., 2018. Development of an influent pollutant disturbance scenario generation using a phenomenological ASM2d-N2O model to describe nitrous oxide emissions in municipal WWTPs under modelling approach. Environ. Model. Softw. 26, 1255–1267. https://doi.org/ dynamic conditions. Chem. Eng. J. 335, 185–196. https://doi.org/10.1016/j. 10.1016/j.envsoft.2011.06.001. cej.2017.10.119. Gernaey, K.V., Jeppsson, U., Vanrolleghem, P.A., Copp, J.B., 2014. Benchmarking of Ni, B.-.J., Yuan, Z., 2015. Recent advances in mathematical modeling of nitrous oxides Control Strategies for Wastewater Treatment Plants. IWA Publishing. https://doi. emissions from wastewater treatment processes. Water Res. 87, 336–346. https:// org/10.2166/9781780401171. doi.org/10.1016/j.watres.2015.09.049. Grau, P., de Gracia, M., Vanrolleghem, P.A., Ayesa, E., 2007. A new plant-wide Nopens, I., Batstone, D.J., Copp, J.B., Jeppsson, U., Volcke, E., Alex, J., Vanrolleghem, P. modelling methodology for WWTPs. Water Res. 41, 4357–4372. https://doi.org/ A., 2009. An ASM/ADM model interface for dynamic plant-wide simulation. Water 10.1016/j.watres.2007.06.019. Res. 43, 1913–1923. https://doi.org/10.1016/j.watres.2009.01.012. Guerrero, J., Guisasola, A., Comas, J., Rodríguez-Roda, I., Baeza, J.A., 2012. Multi- Nopens, I., Benedetti, L., Jeppsson, U., Pons, M.N., Alex, J., Copp, J.B., Gernaey, K.V., criteria selection of optimum WWTP control setpoints based on microbiology-related Rosen, C., Steyer, J.P., Vanrolleghem, P.A., 2010. Benchmark Simulation Model No failures, effluent quality and operating costs. Chem. Eng. J. 188, 23–29. https://doi. 2: finalisation of plant layout and default control strategy. Water Sci. Technol. 62, org/10.1016/j.cej.2012.01.115. 1967–1974. https://doi.org/10.2166/wst.2010.044. Guerrero, J., Guisasola, A., Vilanova, R., Baeza, J.A., 2011. Improving the performance Ostace, G.S., Baeza, J.A., Guerrero, J., Guisasola, A., Cristea, V.M., Agachi, P.Ş., of a WWTP control system by model-based setpoint optimisation. Environ. Model. Lafuente, F.J., 2013. Development and economic assessment of different WWTP Softw. 26, 492–497. https://doi.org/10.1016/j.envsoft.2010.10.012. control strategies for optimal simultaneous removal of carbon, nitrogen and Guisasola, A., Petzet, S., Baeza, J.A., Carrera, J., Lafuente, F.J., 2007. Inorganic carbon phosphorus. Comput. Chem. Eng. 53, 164–177. https://doi.org/10.1016/j. limitations on nitrification: experimental assessment and modelling. Water Res. 41, compchemeng.2013.03.007. 277–286. https://doi.org/10.1016/j.watres.2006.10.030. Otterpohl, R., Freund, M., 1992. Dynamic models for clarifiers of activated sludge plants Gustavsson, D.J.I., Tumlin, S., 2013. Carbon footprints of Scandinavian wastewater with dry and wet weather flows. Water Sci. Technol. 26, 1391–1400. https://doi. treatment plants. Water Sci. Technol. 68, 887–893. https://doi.org/10.2166/ org/10.2166/wst.1992.0582. wst.2013.318. Pocquet, M., Wu, Z., Queinnec, I., Spérandio, M., 2016. A two pathway model for N 2 O Hauduc, H., Rieger, L., Takács, I., Héduit, a, Vanrolleghem, P.A., Gillot, S., 2010. emissions by ammonium oxidizing bacteria supported by the NO/N 2 O variation. A systematic approach for model verification: application on seven published Water Res. 88, 948–959. https://doi.org/10.1016/j.watres.2015.11.029. activated sludge models. Water Sci. Technol. 61, 825–839. https://doi.org/10.2166/ Rieger, L., Alex, J., Winkler, S., Boehler, M., Thomann, M., Siegrist, H., 2003. Progress in wst.2010.898. sensor technology–progress in process control? Part 1: sensor property investigation Hauduc, H., Takács, I., Smith, S., Szabo, A., Murthy, S., Daigger, G.T., Spérandio, M., and classification. Water Sci. Technol. 47, 103–112. https://doi.org/10.2166/ 2015. A dynamic physicochemical model for chemical phosphorus removal. Water wst.2003.0096. Res. 73, 157–170. https://doi.org/10.1016/j.watres.2014.12.053. Santín, I., Barbu, M., Pedret, C., Vilanova, R., 2018. Fuzzy logic for plant-wide control of Hauduc, H., Wadhawan, T., Takács, I., Johnson, B., Bott, C., Ward, M., 2019. biological wastewater treatment process including greenhouse gas emissions. ISA Incorporating sulfur reactions and interactions with iron and phosphorus into a Trans 77, 146–166. https://doi.org/10.1016/j.isatra.2018.04.006. general plant-wide model. Water Sci. Technol. 79, 26–34. https://doi.org/10.2166/ Santín, I., Barbu, M., Pedret, C., Vilanova, R., 2017. Control strategies for nitrous oxide wst.2018.482. emissions reduction on wastewater treatment plants operation. Water Res. 125, Henze, M., Gujer, W., Mino, T., van Loosedrecht, M., 2015. Activated Sludge Models 466–477. https://doi.org/10.1016/j.watres.2017.08.056. ASM1, ASM2, ASM2d and ASM3. Water Intell. Online 5. https://doi.org/10.2166/ Solís, B., Guisasola, A., Pijuan, M., Corominas, L., Baeza, J.A., 2022. Systematic 9781780402369, 9781780402369–9781780402369. calibration of N2O emissions from a full-scale WWTP including a tracer test and a 12 B. Solís et al. Water Research 215 (2022) 118223 global sensitivity approach. Chem. Eng. J. 134733 https://doi.org/10.1016/j. Takács, I., Patry, G.G., Nolasco, D., 1991. A dynamic model of the clarification- cej.2022.134733. thickening process. Water Res. 25, 1263–1271. https://doi.org/10.1016/0043-1354 Solon, K., Flores-Alsina, X., Kazadi Mbamba, C., Ikumi, D., Volcke, E.I.P., (91)90066-Y. Vaneeckhaute, C., Ekama, G.A., Vanrolleghem, P.A., Batstone, D.J., Gernaey, K.V., Torà, J.a, Lafuente, F.J., Baeza, J.A., Carrera, J., 2010. Combined effect of inorganic Jeppsson, U., 2017. Plant-wide modelling of phosphorus transformations in carbon limitation and inhibition by free ammonia and free nitrous acid on ammonia wastewater treatment systems: impacts of control and operational strategies. Water oxidizing bacteria. Bioresour. Technol. 101, 6051–6058. https://doi.org/10.1016/j. Res. 113, 97–110. https://doi.org/10.1016/j.watres.2017.02.007. biortech.2010.03.005. Solon, K., Flores-Alsina, X., Mbamba, C.K., Volcke, E.I.P., Tait, S., Batstone, D., Vaneeckhaute, C., Claeys, F.H.A., Tack, F.M.G., Meers, E., Belia, E., Vanrolleghem, P.A., Gernaey, K.V., Jeppsson, U., 2015. Effects of ionic strength and ion pairing on (plant- 2018. Development, implementation, and validation of a generic nutrient recovery wide) modelling of anaerobic digestion. Water Res. 70, 235–245. https://doi.org/ model (NRM) library. Environ. Model. Softw. 99, 170–209. https://doi.org/ 10.1016/j.watres.2014.11.035. 10.1016/j.envsoft.2017.09.002. Stare, A., Vrečko, D., Hvala, N., Strmčnik, S., 2007. Comparison of control strategies for Vanrolleghem, P.A., Corominas, L., Flores-Alsina, X., 2010. Real-Time Control and nitrogen removal in an activated sludge process in terms of operating costs: a Effluent Ammonia Violations Induced by Return Liquor Overloads. Proc. Water simulation study. Water Res. 41, 2004–2014. https://doi.org/10.1016/j. Environ. Fed. 2010, 7101–7108. https://doi.org/10.2175/193864710798207503. watres.2007.01.029. Wentzel, M.C., Ekama, G.A., Sötemann, S.W., 2006. Mass balance-based plant-wide Stumm, W., Morgan, J.J., 1996. Aquatic Chemistry: Chemical Equilibria and Rates in wastewater treatment plant models - Part 1: biodegradability of wastewater organics Natural Waters, 3rd ed. John Wiley and Sons, New York, NY, USA. under anaerobic conditions. Water SA 32, 269–275. https://doi.org/10.4314/wsa. Su, Q., Domingo-Félez, C., Zhang, Z., Blum, J.M., Jensen, M.M., Smets, B.F., 2019. The v32i3.5261. effect of pH on N2O production in intermittently-fed nitritation reactors. Water Res. Wett, B., Rauch, W., 2003. The role of inorganic carbon limitation in biological nitrogen 156, 223–231. https://doi.org/10.1016/j.watres.2019.03.015. removal of extremely ammonia concentrated wastewater. Water Res. 37, Sweetapple, C., Fu, G., Butler, D., 2015. Does carbon reduction increase sustainability? A 1100–1110. https://doi.org/10.1016/S0043-1354(02)00440-2. study in wastewater treatment. Water Res. 87, 522–530. https://doi.org/10.1016/j. Zhang, W., Wang, D., Jin, Y., 2018. Effects of inorganic carbon on the nitrous oxide watres.2015.06.047. emissions and microbial diversity of an anaerobic ammonia oxidation reactor. Sweetapple, C., Fu, G., Butler, D., 2014. Multi-objective optimisation of wastewater Bioresour. Technol. 250, 124–130. https://doi.org/10.1016/j.biortech.2017.11.027. treatment plant control to reduce greenhouse gas emissions. Water Res. 55C, 52–62. https://doi.org/10.1016/j.watres.2014.02.018. 13