Effect of nitrite, limited reactive settler and plant design configuration on the predicted performance of a simultaneous C/N/P removal WWTP

This paper describes a modelling study where five new benchmark plant design configurations for biological nutrient removal (A(2)/O, UCT, JHB, MUCT and BDP-5 stage) are simulated and evaluated under different model assumptions. The ASM2d including electron dependent decay rates is used as the reference model (A1). The second case (A2) adds nitrite as a new state variable, describing nitrification and denitrification as two-step processes. The third set of models (A3 and A4) considers different reactive settlers types (diffusion-limited/non limited). This study analyses the importance of these new model extensions to correctly describe the nitrification behaviour and the carbon source competition between ordinary heterotrophic organisms (OHO) and polyphosphate accumulating organisms (PAO) under certain operation conditions. The economic and environmental aspects when meeting the P discharge limits by adding an external carbon source are also studied.


INTRODUCTION
Nowadays, water shortage is forcing governments to impose stricter effluent discharge limits for nutrients to wastewater treatment plants (WWTPs). Consequently, upgrading current WWTPs with biological nutrient removal (BNR) including nitrification/denitrification and Enhanced Biological Phosphorus Removal (EBPR) should be a short-term aim. EBPR, the most sustainable technology for phosphorus (P) removal, is based on the enrichment of activated sludge with polyphosphate accumulating organisms (PAO), usually under sequential anaerobic and aerobic or anoxic conditions so that the electron donor (organic matter) and the electron acceptor (usually oxygen but also nitrate or nitrite) are physically separated (Metcalf and Eddy, 2003).
The most widespread mathematical models to describe EBPR in a WWTP are the Activated Sludge Model No. 2d (ASM2d) (Henze et al., 2000) as well as the extended Activated Sludge Model No. 3 (ASM3) incorporating EBPR process (ASM3-BioP) (Rieger et al., 2001). The formulation of these models includes some simplifications to reduce the model complexity and thus, they may not be valid for all scenarios (Sin and Vanrolleghem, 2006). For example the ASM2d default model structure does not differentiate amongst the anaerobic, anoxic and aerobic decay rates while experimental results show the contrary (Nowak et al., 1995;Siegrist et al., 1999). Hence, including nitrite in the ASM models is also essential for achieving a proper description of EBPR in a WWTP. Most of the studies that included nitrite as state variable considered two-step nitrification and denitrification processes. Although the two-step nitrification assumption is commonly accepted, two-step denitrification modelling is not well established and different approaches have been proposed (Wett and Rauch, 2003;Sin and Vanrolleghem, 2006;Guerrero et al., 2011). Sin et al. (2008) analysed some of these models and proposed some guidelines for a consistent description of activated sludge systems including nitrite and considering two-step nitrification and denitrification.
The biological reactions occurring in the secondary settler are another factor to take into consideration when modelling BNR. Although the settling process is usually considered non-reactive (e.g. Takács et al., 1991), several studies (Siegrist et al., 1995) reported that biological reactions also occur, and in particular denitrification processes, despite of the mass transfer limitations present in the settler (concentration gradients and preferential pathways). Gernaey et al. (2005) and Flores-Alsina et al. (2012) presented a reactive settler model that considered each layer of the settler as a continuously stirred tank reactor (CSTR). Unfortunately this approach seems to overestimate the reactive capacity of the settler, since mass transport limitations were not considered. Clearly, more research should be conducted on this topic to correctly simulate a reactive settler.
In a real WWTP, EBPR has to coexist with biological nitrogen (N) removal based on the aerobic nitrification and anoxic denitrification processes. Coupling N removal and EBPR is not just as simple as adding an extra anaerobic zone after the influent inlet to favour PAO growth, since there can be some detrimental interactions between both processes. Most of the reported WWTP configurations for simultaneous N and P removal have an aerobic zone before the secondary settler which may result in the presence of some nitrate or nitrite (named NO X hereafter) in the external recycle (Q EXT ).
The NO X would then enter the anaerobic zone via the external recycle, leading to EBPR failure as reported for many full-scale WWTPs (Henze et al., 2008). The most commonly accepted hypothesis to describe this failure is that the NO X presence triggers the competition for the electron donor (i.e. organic matter) between ordinary heterotrophic organisms (OHO) and PAO. Guerrero et al. (2011) experimentally observed that the nature of the carbon source rules such competition. Thereby, it was proved that OHO were able to outcompete PAO when the electron donor was a complex carbon source, while PAO won the competition when treating a wastewater with high volatile fatty acids content.
Among all the possible WWTP configurations, the A 2 /O (anaerobic-anoxicaerobic) configuration has been widely applied for municipal WWTP despite the obvious disadvantage that complete denitrification is not possible and some NO X will always enter the anaerobic phase via Q EXT (Henze et al., 2008). Thereby, alternative configurations have been designed to prevent such deleterious effect on EBPR by reducing the NO X in the inlet to the anaerobic phase ( Figure 1). The Bardenpho 5-stage (BDP-5 stage) system (Barnard, 1976) improves N removal by adding an extra anoxicaerobic zone and thus, limits the NO X load in the external recycle. Rabinowitz and Marais (1980) designed the UCT (University of Cape Town) system aiming at preventing the Q EXT from entering the anaerobic reactor directly. In this configuration, Q EXT is discharged to the anoxic reactor together with the internal recycle (Q INT ) to denitrify the NO X . A new recycle is then required from the anoxic reactor to the anaerobic reactor to maintain the desired biomass concentration, which is named anaerobic recirculation (Q ANAE ) in this study. However, it has been reported for this configuration (Henze et al., 2008) that avoiding NO X presence in Q ANAE is critical to achieve a high EBPR activity, but this control is not always possible under full scale operation. A modification of the UCT (Modified UCT, MUCT) was proposed to avoid this problem and increase its efficiency. In the MUCT configuration, the Q EXT is directed to an anoxic reactor that does not receive the Q INT flow (Figure 1), easing the total NO X depletion in the Q ANAE . On the other hand, most of the denitrification takes place in the second anoxic tank, which also receives the Q INT recycle flow. Finally, Osborn and Nicholls (1978) proposed another alternative to overcome the negative effect of NO X on EBPR, the Johannesburg process (JHB). Here, an anoxic reactor is located in the Q EXT line so that the NO X in the Q EXT is predenitrified. The electron donor for this process could be either part of the influent (influent bypass, IB) or an external carbon source addition.
Selecting the best of these configurations is not a straightforward issue because many variables affect the overall WWTP performance (i.e. influent characteristics, operational conditions or availability of an external carbon source) and thus, a defined framework is required to compare all the possible scenarios under unbiased conditions. Along this line of thinking, the objective of this paper is to evaluate i) the effect of different model assumptions, and ii) the impact of different WWTP configurations on the performance of EBPR coupled to biological N removal. In order to address the first point, the inclusion of nitrite as state variable and biochemical reactions in the settler (with and without considering mass transfer limitations) were analysed and compared under long-term operation (364 days). On top of that, the previous model assumptions were also applied to the five most common EBPR plant configurations found in fullscale WWTPs. Note that this is the first study where benchmark simulations have been conducted using these new plant configurations. Effluent quality, operational costs and discharge levels were used to evaluate the performance of the different plant configurations.

Mathematical models
In the first step of the study, four different approaches to describe BNR and the settling process were evaluated ( Table 2). The biological kinetic model used to describe BNR in this study was the ASM2d (Henze et al., 2000), similarly to other benchmark studies on proposing new model extensions (Gernaey and Jørgensen, 2004;Flores-Alsina et al., 2012). For the first approach (A1), the ASM2d was extended with electron acceptor dependent decay rates as described by Gernaey and Jørgensen (2004). The secondary settler behaviour was modelled using the 10-layer (non-reactive) settler model of Takács et al. (1991). In the second approach (A2), A1 was modified including nitrite as a new state variable, considering nitrification and denitrification as two-step processes (see Supplementary Information S2 for the complete stoichiometric and kinetic description of the model). Once nitrite is considered, two alternative electron acceptors (nitrate and nitrite) are present for denitrification. Hence, a mixed substrate approach was used similar to the ASM2d mixed substrate implementation for acetate (S A ) versus fermentable COD (S F ) in biological carbon removal processes (i.e. including a SNO 2 /(SNO 2 +SNO 3 ) reduction term in the nitrite degradation rate and a SNO 3 /(SNO 2 +SNO 3 ) term in the nitrate degradation rate) (Sin and Vanrolleghem, 2006). The third approach (A3) aimed to introduce the reactive settler concept to consider biotransformations of both soluble and particulate compounds during the settling process. The full set of equations used in A2 was therefore considered in the settler, where each layer was simulated as a CSTR (Gernaey et al., 2005). However, it is known that this approach results in an overestimation of the reactive capacity of the settler since mass transfer problems or limitations (i.e. concentration gradients or preferential pathways) are not considered. For that reason, a fourth approach (A4) was proposed to describe such settler limitations by adding a reduction factor to the kinetics in the settler. The value of such reduction factor was 0.25, which was determined in order to obtain a denitrifying capacity similar to the one which was experimentally observed in real settlers (See Supplementary Information S3.).
All the simulations were conducted in accordance to benchmarking principles (Jeppsson et al., 2007): 300 days simulation to reach steady state using predefined constant influent data, then 609 days of long term dynamic influent. Only the last 364 days were used for evaluation and comparison purposes. The influent profile was

Operational cost index (OCI):
The OCI (Equation1) was calculated according to the BSM1 guidelines (Alex et al., 2008). Aeration energy (AE), mixing energy (ME), pumping energy (PE), sludge production (SP) and the external carbon source addition (EC), described below, were considered. AE and PE were calculated considering the new approaches described in the BSM2 since aeration was found to play a major role in the OCI and thus, AE has a significant impact on the evaluation process (Nopens et al., 2010).

.3.2. Influent and effluent quality indexes (IQI and EQI)
IQI or EQI (kgPU· d -1 )= 1 1000· total 1 t end t start IQI and EQI (Equation 2) were evaluated similar to Copp (2002) where t total is the total evaluation time and Q j the influent or effluent flow rate. PU X (pollutant units of component X) represents the product between weights β X and the concentration of the considered pollutant at time (t). The weights β X suggested by Gernaey and Jørgensen (2004) were used for IQI and EQI evaluation. However, the fact that the ammonium is more harmful for the environment than nitrate or nitrite (Carmango and Alonso, 2006) was also considered and thus, the weights for total Kjeldahl nitrogen (TKN) and for NO X were changed from 20 to 30 and from 20 to10 respectively to take this effect into account (Nopens et al., 2010). Finally, the weight for total phosphorus (TP) was also increased from 20 to 50 in order to favour those plant configurations or operational conditions that resulted in higher bio-P removal.

Nitrogen removal and EBPR performance under different model assumptions
The results obtained in the LT simulation of the A 2 /O configuration are summarised in figure 2 for the four different model assumptions (  S3.) in order to mimic the denitrification capacity observed in full-scale settlers. When the reduction factor was 0.25, the denitrifying capacity in the settler was around 17% of the TN denitrified in the system. This value is similar to the capacity reported for a real WWTP (Siegrist et al., 1995) and it was therefore kept during all the rest of the simulation study. When the approaches A3 and A4 are compared (non-limited and diffusion limited settler, respectively), less optimistic denitrification rates in the bottom of the clarifier were obtained for scenario A4. As a result, there was a higher P concentration in the effluent since the amount of NO X entering into the anaerobic phase via Q EXT was higher. This reduction in the denitrification process efficiency was also evident in the fact that the TN concentration (mainly NO X ) also increased in the effluent ( Figure 2). On the contrary, a slight improvement of P-removal was still observed for A4 compared to A2 (non-reactive settler). Based on these results, it was concluded that it is important to consider the denitrifying capacity of the settler to properly describe EBPR in BNR. A similar behaviour was observed for the other studied WWTP configurations (see Supplementary Information S4. for further details).

Importance of considering reactive settler under certain operation conditions
To gain more insights about the importance of considering a reactive settler, a new scenario analysis is presented to study its effect on the overall BNR processes. However, the fact of considering each layer as a CSTR resulted in an overestimation of the intensity of the processes occurring in the settler. When A3 was considered, the denitrification capacity of the settler was around 43% of the TN denitrified in the system, which is disproportionate when is compared to the 15% reported in the literature for full-scale settlers (Siegrist et al., 1995).

EBPR behaviour under different plant configurations
Taking the conventional A 2 /O as a reference, this section compares alternative configurations (BDP-5stage, JHB, UCT and MUCT, see figure 1) that have been proposed to minimise the detrimental effect on EBPR of NO X entering the anaerobic phase. Based on the previous results, the inclusion of nitrite in AMS2d and the assumption of a diffusion-limited reactive settler (approach A4) were proved to be necessary to obtain a more realistic description of the BNR processes and thus, this approach was used for these simulations. As was mentioned above, this is the first study where benchmark simulations have been conducted using these plant configurations. However, on the basis of the simulations, it can be concluded that the denitrifying capacity was not fully exploited since ANOX1 was oversized considering the low NO X load originating from Q EXT , whereas ANOX2 was overloaded to denitrify the NO X fed by the Q INT.
In the A 2 /O configuration on the contrary, a lower effluent NO X concentration was observed because both ANOX1 and ANOX2 were used to denitrify the NO X from the Q INT instead of only ANOX2 as occurred in the JHB and MUCT configurations. For example the NO X concentration at the end of the JHB-ANOX2 was 6.68 mg N·L -1 while it was 4.46 mg N·L -1 for A 2 /O-ANOX2. Thus, it can be concluded that MUCT and JHB plants give the PAO a competitive advantage compared to denitrifying bacteria, since more of the influent carbon source was channelled in to the EBPR processes.
The UCT plant showed the highest effluent P concentration ( Figure 5) contrary to what was expected taking into account that it is one of the most often reported configurations used to prevent NO X presence in the anaerobic reactor (Rabinowitz and Marais, 1980;Henze et al., 2008). UCT plant simulation results revealed that total NO X depletion was not achieved at the end of the anoxic phase (5.08 mg N·L -1 ) favouring denitrification instead of P release in the anaerobic reactors. This fact is in agreement with statements made in some engineering manuals (Henze et al., 2008;Metcalf and Eddy, 2003) that pointed out that total anoxic NO X denitrification is critical to achieve high biological P removal in the UCT plant. This issue is tackled by the MUCT, which separates the Q EXT and Q INT inlet points at the expense of decreasing even more the TN removal capacity.
Finally, the BDP 5-stage resulted in a high effluent P ( Figure 5). This could be explained due to the location of ANOX2 in this configuration, which was placed after the AER2 and before AER3 (Figure 1). Thus, the Q INT only fed the ANOX1, resulting in less denitrifying capacity mainly for two reasons: i) a reduction of the anoxic volume to denitrify NO X brought by the Q INT (similar to what occurred for JHB and MUCT); and, ii) the low COD available for denitrification that entered into ANOX2 after the aerobic phase (e.g. NO X concentration only decreased from 12.77 to 10.29 mg N·L -1 in such a reactor). As reported by Van Haandel and Van der Lubbe (2007), the BDP-5 stage configuration can perform well with high P-removal as long as sufficient denitrification is ensured in the second anoxic reactor. Otherwise the system is not capable to prevent nitrate to enter in the anaerobic reactor. Barnard (1976) and Osborn and Nicholls (1978) reported some examples of this problem in a BDP-5 stage pilot plant. To solve this problem, external carbon dosage could be introduced in ANOX2 to ensure sufficiently high COD levels to allow such denitrification.

Effect of carbon addition for the different WWTP configurations
Regarding P removal, the simulation results show that none of the plant configurations met the legal effluent P discharge limit of 1.5 mg P·L -1 ( Figure 5). This is mainly because of the low COD content in the wastewater, and the competition between PAO and OHO for the electron donor in the anaerobic reactor. A conventional solution in real systems is the addition of an external carbon source in the ANAE1 tank in order to provide a supplementary amount of readily biodegradable organic matter for EBPR and denitrifying processes (Olsson et al., 2005). In the last scenario analysis  Figure   6A). When no carbon source was added, effluent TP played a major role in the EQI calculation favouring the MUCT and JHB configurations (black bars). However, when carbon source was added, the effluent phosphorus concentration was drastically reduced and TN caused the main differences in the EQI values (white bars). Thus, in this case JHB and MUCT achieved the highest EQI values ( Figure 6A) due to a higher effluent TN concentration (11.74 and 12.64 mg N·L -1 , respectively). When it comes to OCI criteria, using an external carbon source reduced the effluent pollutant loads but at the expense of an increasing OCI ( Figure 6B). The fact of adding external carbon source implies a cost (Equation 1) and thus, the configurations that required higher carbon addition obviously also resulted in higher OCI values. The external carbon addition also resulted in an increase of the SP ( Figure 6C), which also contributed to increased OCI values due to the considerable sludge processing cost. The BDP-5 stage plant resulted in the highest OCI value not only because more external carbon source addition was required to meet the P limit, but also because it resulted in a higher SP. respectively) would be above the discharge limit.

Practical implications of the study
The list of major (practical) implications extracted from this study can be summarised as follows: • The inclusion of nitrite allows a better description of N removal in systems with low aeration because it could avoid the prediction of potential nitrification failure. Instead of that, partial nitrification to nitrite is described in those cases. Moreover, inclusion of nitrite allows a better accounting of the organic matter needed to denitrify (i.e. denitritation requires less COD than denitratation), which enables a better description and understanding of the competition between PAO and OHO for the carbon source, especially in systems with carbon shortage.
• The reactive settler approach with a diffusion-limitation factor (0.25) allowed a more realistic description of the settling process in terms of biological process rates that can be achieved in settlers. If the assumption of a reactive settler model is not considered, the real denitrifying capacity of the system is not reflected and a wrong EBPR failure could be predicted (anaerobic NO X in the inlet is overestimated). On the contrary, not limiting the reactive settler due to diffusion limitation could result in unrealistically high denitrification rates. In addition, the consideration of reactive settler gains importance in systems with high biomass content because of the higher reactivity of the settler. On the contrary, in systems with low biomass content, only physical processes could be used to simulate settling phenomena.

CONCLUSIONS
The improvement provided by the nitrite inclusion in the ASM2d model was clearly demonstrated, avoiding the prediction of N removal failure in systems with low aeration. Diffusion-limited reactive settler model also allowed a more realistic description of the settling process and thus, the settler reactivity was not overestimated.
Regarding the effect of the plant configurations on biological C/N/P removal, the highest biological P removal was obtained for JHB and MUCT (65% and 55%, respectively). UCT and BDP-5-stage configurations resulted in the lowest TP removal because high amounts of NO X entered the anaerobic zone, favouring OHO denitrification instead of EBPR.

ACKNOWLEDGMENTS
Javier Guerrero is grateful for the grant received from the Spanish government (FPU AP2009-1632). This paper was partially written when Javier Guerrero was a guest     Effluent concentrations obtained for SCA1-A (k L a AER 1 and 2 = 120 d -1 ) and SCA1-B (k L a AER 1 and 2 = 80 d -1 ) in the A 2 /O plant configuration when the nitrification / denitrification are described as single (approach A1, black) or two step processes (approach A2, grey). S NH4 corresponds to ammonium nitrogen, S NO3 to nitrate nitrogen, S NO2 to nitrite nitrogen and S PO4 to orthophosphate phosphorus. Effect of the 25% influent flow rate increase on S O2 , S RBS (S A +S F ), S NH4 , S NO3 , S NO2 and S PO4 for the A 2 /O configuration. The non-reactive secondary settler (A2) is compared with a reactive settler (A3) or a diffusion-limited reactive settler (A4). Black circles correspond to A2, open circles correspond to A3 and grey circles to A4. S O2 corresponds to dissolved oxygen, S RBS to readily biodegradable organic substrates, S NH4 to ammonium nitrogen, S NO3 to nitrate nitrogen, S NO2 to nitrite nitrogen and S PO4 to orthophosphate phosphorus.  Figure 6 Simulations results for the five plant configurations without carbon source addition (black) and when adding an external carbon source to achieve 1.5 mg·L -1 P-PO 4 -3 in the effluent (white).