Clogging limitation of nitrifying biofilters: BiostyrDuo^® process study

The removal of ammonia in wastewater is of critical importance in densely populated areas. In this situation the simple treatment of carbon matter is not enough to prevent the degradation of the receiving water body, as ammonia is an acute toxic substance and also leads to oxygen consumption and ecosystem deterioration. For these reasons, amongst others, the European Union requires its member states to regulate ammonia emissions in surface waters where it is suspected to deteriorate the quality of the receiving body (European Framework Directive 2000).

The biological removal of nitrogen is typically achieved through nitrification and denitrification, during which ammonia is first transformed into nitrate, and then reduced to nitrogen gas, N₂. Such treatment can be performed with a series of anoxic- and aerobic-activated sludge tanks, but it can also be realized with several other types of technologies, including fixed-film reactors. This type of technology allows the nitrifying bacteria to grow on a media that remains inside the reactor, thus partially decoupling the sludge residence time (SRT) from the hydraulic residence time (HRT) and allowing a smaller necessary treatment volume. Among this category of reactors, biofilters have been used for tertiary nitrification in a wide range of operating conditions (Canler et al. 2003; Likso et al. 2004; Rocher et al. 2008; Albuquerque et al. 2012). Biofiltration is, however, not without its flaws. The presence of a short process HRT makes biofilters more subject to sudden changes in pollution loads or operating conditions, as their buffer capacity suffers from the decreased wastewater transit times. The process requires a tight control in order to maintain an effluent quality as constant as possible.

Biofilters combine a biological reactor and a physical filter in a single structure for the removal of the pollution. The ammonia contained in the influent is transformed into nitrate by active biomass, which develops on a submerged media of expanded polystyrene beads called Biostyr^®. The oxygen required for the biological reaction is provided by an upflow of process air, injected co-currently to the water supply. The retention of suspended solids (SS) and excess biomass is carried out by the filtration of the flow through the Biostyr^® beads. That's why biofilters are known to progressively clog during treatment as the filtered particles accumulate and biofilm grows on the media. This clogging reduces the filtration and pollutant removal efficiencies. They thus need to be frequently backwashed in order to remove excess biomass. To do so, part of the treated water is used for counter-current washing, which eliminates excess biological matter and SS retained in the filter bed. The efficiency of this backwashing step is optimized by a cyclic air injection sequence.

Clogging management is generally considered as a critical factor of biofiltration, especially during cold periods and/or during rains events, when high nutrient and particle loads are applied (Rocher et al. 2008). So, increasing the filtration time and reducing the number of backwashes become the key strategy for optimizing nitrification efficiency.

The objective of the work presented here is to verify the influence of a media, named K5, added to the Biostyr^® beads inside a biofilter. With a greater density than Biostyr^® beads, this media stays at the bottom of the biofilter and is supposed to improve operating conditions. Thus, after verifying the operating ranges, the overall performance of both K5-media and reference biofilters have been compared. The comparison is based on the removal efficiency of particles and nutrients (SS, carbon, nitrogen, phosphorus) and the clogging sensitivity using headloss estimation during filtration. In addition to that, based on the Simbio biofiltration mathematical model, the influence of the bacterial activity potentially shifting on the K5-media has been studied.

The 2.5 Mm³ of wastewater disposed of in the Paris conurbation is transported to plants featuring full treatment process trains (Figure 1).

The plants, which are upstream and downstream of Paris, handle flows between 50,000 and 1,700,000 m³.d⁻¹. The present study focused on the Seine Aval plant (SAV) whose process is suitable, under normal conditions, for complete treatment of carbon, nitrogen and phosphorus. The wastewater treatment comprise four main processes (Azimi & Rocher 2017). After pretreatment, (screening, grit removal and oil separation), the effluent is treated biologically (primary settling tank, aeration tanks and secondary clarifiers). Organic content and some of the SS are removed in that step. Subsequent physicochemical phosphate removal occurs in the clarifier. Lastly, removal of nitrogen is carried out by nitrification–denitrification steps on biofilters, before discharge to the Seine River. It is during the nitrification step on biofilters that clogging happens, requiring from operators to bypass a significant part of the influent.

The nitrification step of the SAV plant is equipped with 84 Biostyr^® biofilters (Rocher et al. 2008). These 84 biofilters are arranged in six blocks of 14 each and all biofilters are the same size (173 m²). The work has been performed on one of the blocks (Figure 2).

One biofilter in the block, named the K5-media equipped biofilter, was filled with 0.70 m of K5-media under 3.50 m of Biostyr^® beads. On the same block, a reference biofilter was chosen containing 3.30 m of Biostyr^® beads without K5-media (Figure 2(b)). These two biofilters were chosen because of the similar way they operate. First, they are situated on the same block, which means that they are fed in the same way based on an equal distribution of the water flow and the air supply. Second, the height of the Biostyr^® beads is quite similar for both biofilters; the only difference comes from the K5-media layer in one of them. So it can be assumed that any difference of efficiency found between the biofilters is related to this additional layer.

The influent water quality has been characterized using an automatic sampler placed at the entrance of the block. The effluent quality at the outfall of each biofilter (K5-media equipped and reference) has been measured separately with automatic samplers (Figure 2(a)).

The physical characteristics of the K5-media and the Biostyr^® beads are reported in Table 1.

The density of the K5 media is 20 times higher than that of Biostyr^® beads. So, inside the biofilter, the K5-media stays at the bottom of the Biostyr^® beads during filtration and backwashing. In this way, the additional layer of media, with a specific surface of around 800 m².m⁻³, may serve as support for growing biofilm intended for nutrient removal. This extra layer should also work as a filter, providing an additional barrier to retain SS without any significant influence on the hydraulic headloss because of the K5-media's size of 25 mm, which is six times larger than the Biostyr^® media.

From December 2016 to February 2017, 21 sampling campaigns have been carried out by collecting influent and effluent of both K5-media equipped and reference biofilters. Samples were collected in 10 L plastic bottles using refrigerated 24 h automatic samplers. For each campaign, 100 mL sub-samples were collected every 15 min during 24 hours (from 07:00 to 06:45). Routine wastewater parameters (SS, carbon, nitrogen, phosphorus) were analyzed. As reported in Table 2, these parameters were measured in accordance with French or international standards.

Quantification limits and uncertainty linked to measurement ranges have also been reported.

During the operation of the biofilter, as the filtered particles accumulate and biofilm grows on the media, the biofilters progressively clog. This clogging reduces the filtration and pollutant removal efficiencies. Thus, the biofilters need to be backwashed periodically in order to remove excess biomass. There are two reasons for a biofilter to undergo backwash. The first is based on the filtration time and the second is based on the headloss measurement. In the nitrification step of the SAV plant, the filtration time to start a backwash has been set to 20 hours. While based on the pressure measurement, the headloss has to reach 2.5 mH₂O in the biofilter to start a backwash. So, considering these operating conditions, the SAV biofilters never reach the maximum headloss value and are automatically backwashed once a day. Thus, to be able to compare the influence of the K5 media on the filtration duration, a numerical model of the nitrifying step has been used.

The nitrifying step has also been studied using a model allowing testing of several operating conditions. For that, the SimBio biofiltration model, built in Matlab^® (Mathworks) with the Simulink^® toolbox, was used (Bernier et al. 2014; Bernier et al. 2015). In SimBio, the biofilter hydraulics is modelled as a series of seven Continuously Stirred Tank Reactors (CSTRs) of equal volumes. The presence of media reduces the volume available for liquid flow by a bed porosity factor ε. The biofilm model divides the biofilm in several CSTRs, the thickness of which varies under the effect of filtration, biomass growth and filter backwash. Backwash efficiency is modelled as the removal of a fixed proportion of biofilm thickness in each reactor, using different removal efficiencies for biomass and for other non-biomass particles. Headloss is first computed for a clean filter updated to account for the biofilm's presence. The headloss is computed individually for each CSTR, after which it is summed to obtain the total headloss prediction. For filtration efficiency and headloss computation, the biofilm is considered to be uniformly distributed around the media particles.

The filtration and biological conversion parts, including the effect of aeration, of the biofilter model were previously calibrated on a full year data set (Bernier et al. 2014). As the biofilm thickness increases with the accumulation of filtered particles and biomass growth, it is essential to obtain SS and nutrient predictions close to the measurements at the effluent before working on headloss predictions. In this case, the simulation results for Chemical Oxygen Demand (COD), NH₄⁺, NO₃⁻ and PO₄³⁻ were reasonably good. The daily variations in SS concentration of the effluent during simulation were a little less accurate, but the scale of removal remained correct. The estimated sludge production masses were used to grossly calibrate the backwash sub-model as no direct measurements of the sludge removal were available. Once the backwash removal efficiency was properly set, the model was calibrated on the post-backwash headloss data. In this case, biofilters were considered to be backwashed once a day.

Table 3 shows the operating conditions (water and air flow) for both biofilters during the three-month period of the study.

The operating conditions of k5-media equipped and reference biofilters were quite similar. Considering the filtration time, K5-media equipped and reference biofilters median values were 21.5 h.d⁻¹ and 21.3 h.d⁻¹, respectively. Moreover, in both cases, the dispersion of values, given by the c90/c10 ratio, has been calculated equal to be 1.3. These results show that both biofilters have the same filtration time and, at least, one backwash per day.

During the filtration period, the water and air feeding flows were in the same order of magnitude. The water rising velocity median values were 4.8 m.h⁻¹ and 4.7 m.h⁻¹ and the air flow median values were 49,968 Nm³.d⁻¹ and 51,576 Nm³.d⁻¹ for the K5-media equipped and reference biofilters respectively. However, since air flow control is based on one ammonium sensor placed at the beginning of the block, it is logical to have the same range values of air flows. To make sure that the operating conditions are similar, the applied loads also have to be compared.

Pollutant loads (ammonium [N-NH₄⁺], SS [kg], COD [kg O₂], and soluble COD [kg O₂]) have been measured to complete the verification of the operating conditions between both biofilters (Figure 3).

For each figure, green bars show the applied loads (kg.d⁻¹) on all biofilters of the block for each of the 24 h-period sampling campaigns while red and blue plots show applied loads on K5-media equipped and reference biofilters related to their layer of material, respectively. Looking at the applied loads on the whole block, pollutants did not have the same behavior considering soluble or particular pollutants. Ammonium and soluble organic matters (CODs) loads ranged from 504 to 1,199 kg N-NH₄⁺.d⁻¹ and 433 to 1,431 kg O₂.d⁻¹ respectively and their variation expressed as percentile 90/percentile 10 ratio was 1.5 and 1.9. On the other hand, suspended solid and organic matter (COD) loads ranged from 350 to 1,537 kg.d⁻¹ and from 848 to 2,424 kg O2.d⁻¹ with a percentile 90/percentile 10 ratio of 2.7 and 2.3 respectively. Applied loads on each biofilter are logically in line with these results showing low variation for soluble pollutants and high variation for particular pollutants. Focusing on ammonium, mean (±SD) loads of 1.04 (±0.22) kg N-NH₄⁺.m⁻³.d⁻¹ and 1.31 (±0.28) kg N-NH₄⁺.m⁻³.d⁻¹ have been calculated for K5-media equipped and reference biofilters, respectively. Since load values have been calculated taking into account the whole material in biofilters, values for the K5-equipped biofilter are continually less than those of the reference; the latter having less material (3.30 m) than the K5-media equipped one (4.20 m). Rocher et al. (2012) reported an optimal operating conditions of approximatively 1.1–1.2 kg N-NH₄⁺ per m³ of Biostyr^® media and per day, which means that applied loads on biofilters during this study may be considered as normal and usual operating conditions.

The aeration rate, which is the air flow injected per kg of removed ammonium for each sampling 24 h-period, were also calculated. The mean (±SD) values were 84 (±20) Nm³.kg⁻¹_{(N-NH4 removed)} and 83 (±19) Nm³.kg⁻¹_{(N-NH4 removed)} for K5-media equipped and reference biofilters, respectively. No significant difference has been shown using the Student test (α = 0.01). Thus, during tests, both biofilters were fed in the same way with an equal distribution of water, air and pollutant loads.

Considering these operating conditions, the influence of the K5-media on the efficiency of the nitrification step has been evaluated comparing the overall performance and the clogging sensitivity of both biofilters.

The overall removal efficiencies of both biofilters were similar. Considering specifically nitrogen treatment, mean ammonium removal rate calculated for the whole sampling campaigns was 88.0 and 88.4% for K5-media equipped and reference biofilters, respectively. Not surprisingly, the nitrate production yields were also similar with 1,850 and 1,927%. The slight difference observed for SS with a removal efficiency of 61.7% and 67.1% for K5-media equipped and reference biofilters was not statistically relevant. The residual output concentrations were also similar between both biofilters and close to the output set values.

The comparison of biofilter removal efficiencies had also been done taking into account the applied loads. Figure 4 presents removal loads from K5-media equipped biofilter versus the reference one for each sampling campaign.

Generally, it could be said that both biofilters have the same capacity for pollutant removal since all plots are close to the y = x line. However, concerning soluble pollutants (ammonium and soluble COD), the removal loads from the K5-media equipped biofilter was strongly correlated to those with the reference biofilter. The coefficient of determination are 0.99 and 0.97 for ammonium and soluble COD, respectively. In the ammonium case, an increase of the removal loads with the K5-media equipped biofilter is observed. When removal loads are higher than 700 kg N-NH₄⁺.d⁻¹, K5-media equipped biofilter has a higher estimated removal capacity than the reference biofilter. This additional removal capacity has been estimated at around 7% for high ammonium applied loads. Concerning SS and COD (particular matter), the plots are more dispersed and the coefficient of determination are lower, equaling 0.76 and 0.79, respectively. No distinction for the K5-media equipped or reference biofilter has been observed.

Both biofilters have exactly the same capacity to remove particulate and dissolved pollutants under usual conditions. In the ammonium case and under specific conditions of high applied loads, the removal efficiency looks better by having a K5-media layer. This result suggests that the additional media may contribute to pollutant removal by developing a biofilm layer on the K5-media. In the case of particular matters, the K5-media does not seem to play a major role for their removal.

As with Biostyr^® beads, the K5-media seems to serve as a biofilm support. So, bacterial growth influence on the biofilter headloss has to be evaluated. Figure 5 shows the comparison of initial headloss for the K5-media equipped and the reference biofilters. The initial headloss, measured after each backwash, allows us to evaluate the biofilter clogging and the backwash ability to remove excess biomass. The first month of data has been discarded to ensure that biofilm has developed on the K5-media.

The initial headloss has been normalized with the volume of beads and K5-media and with the water rising velocity. It varies from 1 to 3.6 cm H₂O.m⁻³.m.h⁻¹ for the K5-media equipped biofilter (Figure 5(a)). For the reference biofilter, excluding two results under 1 cm H₂O.m⁻³.m.h⁻¹, the initial headloss ranged from 2.3 to 6.1 cm H₂O.m⁻³.m.h⁻¹ and is always above those of the K5-media equipped biofilter. Figure 5(b) shows the same results under boxplot form. This form, introduced by Tukey (1977) consists of a box extending from the first quartile (Q1) to the third quartile (Q3); a bar mark at the median and a cross mark at the mean; and whiskers. The schematic boxplot divides the data based on four invisible boundaries, namely, two inner fences and two outer fences. As usual, the interquartile range (IQR) is defined to be Q3–Q1. The inner fences are Q1–1.5 IQR and Q3 + 1.5 IQR, while the outer fences are Q1–3 IQR and Q3 + 3 IQR. The whiskers extend to the most extreme data within the inner fences. Data outside the outer fences are considered to be extreme outliers and are marked with a symbol. This statistical approach shows that values under 1 cm H₂O.m⁻¹.m.h⁻¹ are outliers. Mean initial headloss values are 2.2 and 4.0 cm H₂O.m⁻³.m.h⁻¹ for the K5-media equipped and the reference biofilters, respectively. Moreover, the Mann Whitney tests confirm that initial headloss of the K5-media equipped biofilter is lower than the reference one. This result shows that the 0.70 m of K5-media does not introduce additional headloss into the biofilters under normal operating conditions while it plays a role of a usual fixed bed, as Biostyr^® beads, allowing bacterial growth. Since a part of the pollution is removed before reaching the Biostyr^® beads, the headloss within the Biostyr^® beads is lower than in the reference biofilter. It can also be assumed that the K5-media may play a role on the air transfer inside the media, helping to reduce the total amount of air blown inside the biofilters. So, knowing that air blowing is one of the most expensive processes in a WWTP, K5-media may also help to reduce operation costs of the nitrification step.

As described above, both biofilters had backwash at least once a day. There are two reasons for a biofilter to undergo backwash. The first is based on the filtration time and the second is based on the headloss measurement. Since in our case, it is because of the first reason that both biofilters undergo backwashes, their comparison does not allow evaluation of the influence of the K5-media on the filtration time. To do so, the SimBio model has been used for three steps. First, its capacity to predict the NH₄⁺ residual values and the headloss of the reference biofilter has been checked. Second, the simulations obtained by the model have been adapted to the K5-media biofilter and compared to the reference ones to confirm the headloss gain brought by the added media. Finally, the filtration time predictions obtained with both biofilters have been compared.

• Nitrification and headloss modelling with reference biofilter

The model results for the reference biofilter on the month-long January 2017 dataset are presented in Figure 6 for NH₄⁺ residual and headloss.

Simulation results for the hourly ammonia concentration in the effluent of the biofiltration lane containing the reference biofilter are compared to the measurements made onsite at the same location in Figure 6(a). Although the model initially underestimates the observed concentrations for the first five days, simulation results remain close to the measurements for most of the remaining period. This initial underestimation is apparently caused by an imperfect set of initial conditions following the steady state simulation used to estimate them. It means that the model slightly overestimates the available nitrifying biomass in the biofilter. The mean error (ME) and mean average error (MAE) are in this case −1.14 and 2.12 mgN/L (n = 745), respectively, indicating as noted a slight overestimation of the nitrifying capacities of the simulated biofilter. These results are of similar order as those obtained during the initial model calibration and validation work on the same biofiltration process (Bernier et al. 2014).

Simulation results for the hourly headloss in the reference biofilter are also compared to the measurements made onsite at the same location in Figure 6(b). The observed headloss evolution, both in the case of important short-term variations (such as peaks found around days 10, 12 and 21–25) as well as for the global headloss level increases and decreases across the month, are generally simulated well. Short-term variations are generally caused by sharp increases in the influent flowrate due to rain events and the associated increase in water volumes reaching the Seine Aval WWTP. Mid to long-term evolutions are caused by the gradual accumulation or loss of biofilm and biofilm thickness following the different ammonia loads across the simulated periods. Some specific headloss data (day 11) are simulated with slightly less precision, as well as data points that are particularly high or low. Nevertheless, the ME and MAE scores remain low: 0.04 and 0.13 mH₂O respectively (n = 692). Once again, these results are similar to those obtained during prior work on the subject (Bernier et al. 2015). The model is thus able to reliably simulate the nitrification behavior of the reference biofilter, as well as the headloss resulting from it.

• SimBio adaptation to K5-media

To adapt SimBio to the K5-media equipped biofilter, it has been assumed that because of the bacterial growth on its surface, the additional media allows the removal of most of the dissolved organic matter and a small part of the particulate matter (less than 25%) admitted on the Biostyr^® beads of the nitrification stage. In this way, Figure 7 shows ammonia and headloss simulation results for the K5-media equipped biofilter on the month-long January 2017 dataset compared to simulation values on the reference biofilter.

The simulation of the NH₄ concentration in the effluent is slightly higher for reference biofilter than for K5-media equipped biofilter irrespective of the residual level. The median value of NH₄⁺ residual is 2.18 mg N.L⁻¹ and 0.77 mg N.L⁻¹ for the reference and the K5-media equipped simulations, respectively.

The headloss values simulated for the K5-media equipped biofilter are significantly below those of the reference biofilter. The median values of 0.45 mH₂O and 0.74 mH₂O for respectively the K5-media equipped and the reference are linked to the fact that bacterial growth involved in the residual COD removal is made on K5-media, which favorably influences headloss. Therefore, the bacterial growth on the Biostyr^® beads is less important, reducing the associated headloss. However, because of the method used to adapt the SimBio model, it is possible that an additional headloss brought on by the K5-media has been neglected.

• Filtration time prediction

As said above, in this study, it is because of a filtration time that both biofilters undergo backwashes. So, to make possible the comparison of the filtration time with or without the K5-media, a maximum headloss value of 0.8 mH₂O has been set in the Simbio model to fix backwash condition. Table 4 presents the simulated number of backwashes for the reference and the K5-media equipped biofilters during the January 2017 time-period.

In this condition, the number of backwashes for the reference biofilter is 1.16 (20:40 hours of filtration before backwash), which is quite similar to the numbers of backwashes seen on the plant (median value of 21:30 hours of filtration, Table 3). Considering the K5-media equipped biofilter, for the same backwash condition set at 0.8 mH₂O, the number of backwash decreases to 0.74 (32:20 hours of filtration before backwash). So, using the K5-media allows an increase of 56% of the filtration time before backwash. However, this value is most probably an overestimation since the headloss causes by the K5-media is not simulated by SimBio. It can be assumed that the increase of the filtration time using the K5-media may reduce the operation costs linked to the backwash water pumping and treatment.

Clogging management is considered a key factor of biofiltration, especially when high nutrient and particle loads are applied. So, the main goal of operators is to limit the headloss inside biofilters while keeping the longest possible filtration time and lowest number of backwashes. To do so, one solution may come from the addition of a layer of media allowing the increase of biofilter removal capacity without influencing headloss. The comparison of a biofilter equipped with layer of a K5-media to a reference biofilter showed a similar overall performance under normal operating conditions of the WWTP. The clogging sensitivity of the tested biofilter was different. The initial headloss measured in the K5-media equipped biofilter was 45% lower than the reference one. This result, obtained by measurements, has been confirmed by simulations performed on the validated and calibrated SimBio modelling tools. The headloss during filtration simulated for the K5-media equipped biofilter has been estimated at 40% lower than for the reference biofilter. Even if this value is most probably an overestimation, it can be assumed that this difference is attributed to the bacterial growth on the K5-media surface. This bacterial activity participates in nutrient removal and allows the increase of the nitrification biofilter's performance, mostly for high loads. While the shape of the K5-media does not induce additional headloss. Moreover, the K5-media allows a significant increase of filtration time of more than 50%. So, the operation costs linked to the backwash water pumping and treatment may be reduced using such devices on existing plants.