Introduction

The recast of the European Urban Wastewater Treatment Directive (UWWTD)1 requires the implementation of quaternary treatment stages to remove organic micropollutants (OMPs) in many municipal wastewater treatment plants (WWTPs) across Europe. Membrane hybrid processes that combine ultrafiltration (UF) and adsorption onto powdered activated carbon (PAC) can effectively remove OMPs from municipal wastewater2,3,4. At the same time, membrane hybrid processes create a water quality that is suitable for further beneficial uses, such as bathing water quality or non-potable water reuse (according to EU 2020/7415), with little to no additional treatment2,6,7. However, membrane hybrid processes are not yet widely applied in municipal WWTPs. The potential reasons for this include a lack of full-scale references for this use case and a relatively young history of economic feasibility of membrane filtration in municipal wastewater treatment8.

UF membranes (pore diameter of ~0.02 µm) ensure complete removal of PAC, even for fine products (median particle diameter in the low µm range)9. Such fine PAC products have shown superior adsorption properties, such as accelerated adsorption kinetics and slightly higher adsorption capacity for some compounds, as opposed to conventionally sized PAC10,11,12,13. These improved properties may reduce the required adsorption contact time and, hence, the necessary reactor volumes. Novel inline-dosing process designs with fine PAC entirely skip dedicated contact reactors. The result is a significantly reduced contact time of PAC and water in the membrane stage feed pipe of only several seconds (cf. t1 in Fig. 1a). Additionally, in such setups, PAC is available for adsorption while it accumulates in the membrane cake layer over the course of a filtration cycle (cf. t2 in Fig. 1).

Fig. 1: Simplified process scheme with counter-current principle owing to PAC recirculation.
Fig. 1: Simplified process scheme with counter-current principle owing to PAC recirculation.
Full size image

a Simplified process scheme of the investigated CASP & PAC + UF membrane hybrid process with recirculation of partially loaded PAC to the CASP, and b simplified PAC counter-current principle. Including indication of relevant PAC contact volumes in their respective location within the treatment train: t1 refers to the hydraulic contact time within the connecting pipework between feed pump and membrane module; t2 is the average solid retention time in the filter cake layer of the membrane; t3 is the solid retention time of recirculated PAC in the CASP. A more detailed flowchart of the investigated process is shown in Fig. 6 of the Materials and Methods section.

However, inline-dosing processes have struggled to achieve 80% OMP removal with PAC dosages comparable to those used in more established PAC adsorption stage designs. Presumably, in inline-dosing processes, the applied contact times were insufficient to fully load the PAC, despite the use of fine PAC products2,6. To overcome this challenge, optimal operating conditions and improved carbon utilisation in inline-dosing membrane hybrid processes have recently been investigated, as they are key factors to economic feasibility. For example, Hoffmann et al. and Schwaller et al.3,4 have investigated optimal dosing schemes for PAC in the feed pipe and in the membrane cake layer, thereby improving PAC utilisation in the PAC + UF stage. The approach in this present study is rather to recirculate partially loaded PAC from the PAC + UF stage to the upstream conventional activated sludge process (CASP), applying a counter-current principle of PAC and water (and, thus, OMP concentration), as a possible solution to better utilise the entire adsorption capacity of the PAC (cf. Fig. 1b).

Through recirculation, PAC remains in the wastewater treatment process until it is removed as part of the CASP’s excess sludge. The resulting contact time (cf. t3 in Fig. 1) is, therefore, closely connected to the sludge age, which commonly exceeds practically relevant PAC adsorption equilibrium times of 48 hours14. Based on modelling15, lab experiments14 and pilot-scale experiments16, it has been proposed that the PAC dosage may be reduced by up to 50% through PAC recirculation, where most of the literature focuses on recirculation within the adsorption stage.

To the best of our knowledge, this study is the first to investigate the effect of PAC recirculation on the upstream CASP from an inline-dosing membrane hybrid process at pilot scale. The investigation of effects on OMP removal was possible because the treatment capacities of the available CASP and PAC + UF processes (~0.5 m3/h inflow) matched. In this study, we specifically seek to answer the following questions:

  1. (i)

    How much can the PAC dosage be reduced under real conditions when implementing the recirculation of partially loaded PAC from the UF stage to the upstream CASP?

  2. (ii)

    Where can the additional adsorption of OMPs be localised?

  3. (iii)

    What mechanisms does PAC recirculation trigger that affect treatment performance?

Results and discussion

Recirculation of PAC reduces dosage requirements in inline-dosing membrane hybrid processes by up to 50%

In the two observed inline-dosing membrane hybrid processes, recirculating partially loaded PAC to the CASP stage improved OMP removal (see Fig. 2a, b). In the inline-dosing process with fine PAC (median diameter D50 = 5 µm), the required DOC-specific dosage for an 80% removal of OMPs was reduced from an initial 1.4 mgPAC/mgDOC without recirculation down to <0.7 mgPAC/mgDOC with recirculation (see Fig. 2a). The measured DOC concentration in the CASP effluent served as a reference for calculating specific dosages across all settings.

Fig. 2: Effect of PAC recirculation on the required DOC-specific PAC dosage for 80% OMP removal (in accordance with UWWTD).
Fig. 2: Effect of PAC recirculation on the required DOC-specific PAC dosage for 80% OMP removal (in accordance with UWWTD).
Full size image

OMP removal is shown for two UWWTD-conform substance selections (EU best mix (a, b, c) and EU weighted (d, e, f); see the methods section for the underlying substances) and for three process designs of a pilot-scale CASP & UF + PAC adsorption membrane hybrid process. Removal of the processes without PAC recirculation is based on data from Zimmermann et al.2. The shown data represent the different process configurations: without PAC recirculation (white-filled circles; linear fit by black dotted lines), and with PAC recirculation (pink-filled circles; linear fit by pink solid lines). The panels allow to compare the inline-dosing process with fine PAC (a, d), the inline-dosing process with conventional PAC (b, e), and the Ulm process with conventional PAC (c, f).

Similar improvement through recirculation was observed for the application of conventional PAC (D50 = 30 µm) in the inline-dosing process; however, an overall higher dosage requirement of ~1.5–2.0 mgPAC/mgDOC (cf. Fig. 2b) was required for 80% OMP removal. This dosage requirement range is similar to many conventional process schemes, such as simultaneous dosing into the CASP17 or immediately upstream of fixed-bed filters18,19.

As a reference, we also operated a conventional PAC stage with a separate contact reactor and an internal PAC concentration through a subsequent lamella clarifier. In Germany, this setup is referred to as the “Ulm process” (for further details, see the methods section). The Ulm process showed only very slight to no improvement through the introduction of PAC recirculation to the CASP stage, with a required PAC dosage of 1 mgPAC/mgDOC, which aligns with what the literature describes for this process18. Since this process is designed to fully load the PAC within the adsorption stage itself, no further loading after recirculation was expected when we included the PAC recirculation in the CASP. Slight improvements may be attributable to other impact factors, which are discussed later in this paper. We summarise that recirculation of PAC can lead to removal performance for OMPs improved by a factor of two, which is close to literature predictions14,15.

Counter-current flow unlocks elevated loading of PAC with OMPs

The above observations raise two questions: i) How is the removal performance improved so substantially? and ii) What mechanism leads to lower dosage requirements in the inline-dosing process with fine PAC and recirculation than what is required in the reference Ulm process? We hypothesise that the major improvement in adsorption performance achieved through PAC recirculation in the inline-dosing process can be explained by PAC adsorption kinetics and the available contact times in the system, as discussed in the next section.

The improvement of the required PAC dosage compared to the reference dosage in the Ulm process, however, cannot be explained solely by kinetics. We assume the following reasons majorly contribute to this phenomenon:

The recirculation of PAC to the upstream CASP introduces a counter-current principle to the application of the adsorbent in relation to the OMP concentration, and, at the same time, to the concentration of the competing background matrix in the water phase (cf. Fig. 1, bottom). Upon initial introduction into the process (cf. Fig. 1, t1 + t2), fresh, unloaded PAC adsorbs OMPs from the secondary effluent. This secondary effluent underwent biological treatment and, in the case of PAC recirculation, also underwent adsorption onto recirculated PAC. Hence, competing adsorption in the secondary effluent is minimal. Once separated by the membrane, partially loaded PAC is recirculated to the CASP at the end of a filtration cycle (see SI Fig. S4). In the CASP, the remaining adsorption capacity can be utilised while facing higher OMP concentrations, but also higher competing background concentrations.

Such counter flow principles are well-known and widely applied, for example, in heat exchangers (heat transfer) and other two-phase reactors (mass transfer), leading to an overall increase in average driving force opposed to in co-current systems20.

Other effects that might elevate OMP removal in the inline-dosing process with fine PAC:

i) Potentially, the additional dosing of coagulant into the contact reactor of the Ulm process may lead to hindered availability of adsorption capacity21, e.g., due to pore blockage or longer diffusion pathways to the adsorption surfaces. However, the literature does not fully agree on whether coagulation does or does not interfere with adsorption22. The inline-dosing process offers not only the potential advantage of improved adsorption performance due to reduced coagulant hindrance but also a cost advantage because it does not require additional coagulant dosage in a contact reactor (cf. Fig. 6 in “Methods”).

ii) Interestingly, the counter-current flow of conventional PAC in the inline-dosing process did not outcompete the Ulm process but instead, led to slightly higher required PAC dosages (cf. Fig. 2b; 1.5–2.0 mgPAC/mgDOC vs. Fig. 2c ~ 1 mgPAC/mgDOC). This underlines the advantage of fine PAC in settings with very short contact times. Additionally, this observation may be attributed to the pronounced competitive adsorption of bulk organics in the CASP, which minimised additional adsorption of OMPs onto only slightly loaded, recirculated conventional PAC from the inline-dosing process.

iii) Factors that are difficult to investigate are the mutual effects of PAC recirculation and biological degradation processes. Recirculated PAC can serve as nuclei for sludge formation (cf. SI Fig. S5) and may improve overall sludge properties23. It may also enable slightly improved biological OMP degradation of adsorbed compounds. Through PAC recirculation, these are kept in the process (and, hence, available for biodegradation) for longer than they would without recirculation of PAC16.

Overall, only the observations regarding the consequences of kinetics (see the following section) are considered certain, whereas the improvement compared to the reference system is yet to be fully understood.

PAC recirculation enables reaching maximum loading

The removal improvement through PAC recirculation seems to depend mainly on two factors: i) the PAC type within the inline-dosing process (fine or conventional) and ii) process design (inline-dosing or Ulm process). Both determine how much the PAC can be loaded within the adsorption stage. Since in this study PAC type is differentiated only by size, the major difference can be attributed to the kinetic behaviour of differently sized PAC. Bonvin et al. and others report faster adsorption kinetics for finer PAC6,11,24,25. At the same time, the process designs (inline-dosing or Ulm process) primarily differ in the available contact time of PAC and water. This amplifies the effect of different adsorption kinetics24.

To better understand the kinetic behaviour behind the processes, we performed laboratory adsorption kinetics experiments with the two PACs from the pilot study (cf. Fig. 3), utilising UV254 as a surrogate parameter for micropollutant removal. UV254 has been reported multiple times as a suitable surrogate parameter for OMP removal from municipal wastewater through PAC26,27,28 and is therefore considered suitable to create a basic understanding of the kinetic behaviour of such adsorption processes.

Fig. 3: Kinetic UV254 adsorption behaviour of the conventional and fine product variant of the investigated PAC.
Fig. 3: Kinetic UV254 adsorption behaviour of the conventional and fine product variant of the investigated PAC.
Full size image

a Absolute measured UV254 concentration in dependence of contact time; b calculated UV254 loading on PAC in dependence of contact time; c normalised loading of PAC in dependence of contact time, where 100% is defined as the maximum loading of fine PAC after 48 hours. Conventional PAC (D50 = 30 µm) is shown as filled black squares; fine PAC (D50 = 5 µm) is shown as open circles in all panels. Fitted lines in (a, b) were obtained by applying the Linear Driving Force (LDF) model described in ref. 32, deriving the respective k* kinetic constants (see SI Text S1 for more details on model fit).

As expected, fine PAC reduced UV254 more quickly (cf. Fig. 3a) and, thus, approached its maximum loading much quicker than conventional PAC (cf. Fig. 3b). Additionally, the fine PAC showed a slightly higher final loading after 48 hours (cf. Fig. 3b). This may be explained by two reasons: the conventional PAC did not reach its full loading after 48 hours, or, as some literature suggests, the maximum loading of finer PAC is slightly higher than that of conventional PAC. The reason for the latter is a larger adsorption surface, since previously inaccessible pores are made available29. This leads to a relative reduction of pore blockage effects during the adsorption process13.

In the inline-dosing process, PAC is available for adsorption for 30–60 seconds of hydraulic retention time in the pipeline (cf. Fig. 3c t1) plus 20 minutes of average solids retention time in the cake layer (t2). The adsorption kinetics experiment shows that t1 + t2 is only enough time to reach up to 60% of the maximum loading (here, the loading after 48 hours). For a conventional PAC, only ~45% of its maximum loading could be achieved within the same time (t1 + t2). This is based on the assumption that PAC is as available for adsorption when fixed in the cake layer as it is in the turbulent flow within the pipeline. The exact behaviour of PAC adsorption in membrane cake layers is yet to be fully understood3. We can, however, conclude that PAC cannot be fully loaded in the available contact times in this process design. Hence, when a partially loaded PAC is recirculated to the CASP stage, it contains unoccupied adsorption sites that become available for further loading.

Consistent with this, the Ulm process showed no significant improvement in OMP removal with recirculation of PAC to the CASP. This aligns with the observed kinetic behaviour of PAC, which reached its maximum loading after 48 hours. The contact time in the contact reactor of the Ulm process (including PAC slurry concentration via internal recirculation in a sedimentation stage) is designed to exceed 48 hours to fully load PAC in this stage, as described by ref. 16.

These results also align with predictions based on modelling15 or laboratory experiments14. Both propose a potential reduction in PAC dosage by approximately a factor of two through the introduction of PAC recirculation.

PAC recirculation shifts the reduction of OMPs towards CASP

As discussed above, the PAC recirculation leads to an improved OMP removal in the CASP stage for several potential reasons. This improved removal during the CASP stage was observed for all measured substances; however, the magnitude of the effect differed among compounds. The displayed substances in Fig. 4 were selected due to their high abundance (microgram per litre concentration range) in the CASP influent and according to their properties: i.e., biodegradable and well adsorbing (benzotriazole and sulfamethoxazole), not well biodegradable but adsorbing (diclofenac), neither well biodegradable nor well adsorbing (candesartan).

Fig. 4: Localisation of the (adsorptive and biological) removal of selected OMPs and bulk organics without vs. with PAC recirculation.
Fig. 4: Localisation of the (adsorptive and biological) removal of selected OMPs and bulk organics without vs. with PAC recirculation.
Full size image

Recirculation of fine PAC from the UF stage to the CASP enhanced removal by better utilisation of PAC adsorption capacity compared with a sequential process (CASP then PAC + UF) without PAC recirculation. The applied PAC dosage was 10 mg/L in both cases, with a contact time of 60 seconds in the pipe and, on average, 20 min of solid retention time in the membrane cake layer. Data originates from setting I with PAC recirculation (this study) and from setting 1 without PAC recirculation (Zimmermann et al.2). For further details on the underlying settings, see supplementary material. The different stages and process configurations are colour-coded: CASP stage—without PAC recirculation (green circles), PAC + UF stage—without PAC recirculation (blue diamonds), and process configuration with PAC recirculation (pink circles). The lines shall guide the eye. Points show the average of n = 3 samplings of this one selected setting, whiskers indicate min and max. The average absolute concentration in the plant influent and its standard deviation for each substance are indicated as c0.

With PAC recirculation, easily adsorbable substances such as benzotriazole were reduced to such a great extent within the CASP that diminished influent concentrations limited their absolute removal in the subsequent PAC + UF stage. With PAC recirculation, the inline-dosing process reduced benzotriazole from an average of 6482 ng/L in the CASP influent to below 240 ng/L in the overall effluent at the setting shown in Fig. 4, corresponding to a 96% removal. Overall, diclofenac removal improved, and the substance was reduced within both stages of the treatment train; while without recirculation, no reduction was observed in the CASP stage. Over the entire treatment train, diclofenac removal was improved from ~50% without PAC recirculation to more than 75% with PAC recirculation. The largest improvement in removal was observed for candesartan, which showed only minor removal without recirculation but could be removed by more than 60% with recirculation. On the one hand, this improvement appears large as this substance showed a heavily varying removal performance in the process without recirculation, including an apparent negative removal in the CASP. On the other hand, more difficult to adsorb substances are expected to benefit most from the extended contact time of PAC and water, as slower adsorption kinetics can unfold and potential synergistic effects can occur, as described by Meinel et al.16 and as proposed in previous sections. Sulfamethoxazole, which is known to exhibit variable removal in adsorption processes (due to weak adsorptive binding and potential metabolite transformation in biological processes), showed consistent results in this study. An attempt to correlate the improvement by PAC recirculation to selected physio-chemical OMP properties showed no strong correlation (cf. SI Figs. S6 and S7) and would need further investigation.

Reduction of bulk organics is barely impacted by PAC recirculation, but composition changes

A closer look at absolute DOC and UV254 concentrations along the treatment trains is provided in the supplementary material (see SI Fig. S3). In contrast to the OMPs, DOC removal did not improve significantly with PAC recirculation (see SI Text S4 for statistical analysis). In the context of major improvements in OMP removal through PAC recirculation, this may indicate a degree of selectivity of the adsorption process toward OMPs relative to DOC, possibly due to differences in average molecular size. Micropollutants with small molecular size exhibit higher pore diffusion coefficients and thus benefit more from extended contact times as compared to larger molecules such as the average DOC compound30. This relative advantage of substances with smaller molecular size is amplified by the counter-current of PAC and water (promoting preferential adsorption in both stages—UF and CASP—individually) because of the re-initialised substance concentrations and thus restored diffusive driving force in each stage. This preferential OMP adsorption is favourable for efficient PAC loading, and the evidence also supports the hypothesis that OMPs are not readily displaced by competitive adsorption after recirculation of PAC into the CASP. No changes in DOC concentrations also meant no influence on the reference DOC in the effluent of the CASP, which is used to calculate specific PAC dosage.

The UV254 concentration in the CASP effluent, on the other hand, changed slightly due to the recirculation of PAC to the CASP (see Fig. S3 and statistical tests in SI Text S4). Since UV254 is often used as a surrogate for OMP removal in process control26, an understanding of its behaviour in recirculation setups is important and, so far, has only been examined in laboratory tests27. Therefore, in Fig. 5, we examined the correlations between UV254 reduction (ΔUV254) and OMP removal in both the inline-dosing process and the Ulm process. Corresponding results of statistical tests can be found in SI Table S2.

Fig. 5: Changes in water composition through PAC recirculation to CASP, displayed as the relation between OMP removal (EU weighted) and UV254 removal (i.e., ΔUV254) for the two processes and their CASP and PAC + UF stages.
Fig. 5: Changes in water composition through PAC recirculation to CASP, displayed as the relation between OMP removal (EU weighted) and UV254 removal (i.e., ΔUV254) for the two processes and their CASP and PAC + UF stages.
Full size image

The reference UV254 (and therefore underlying absolute UV254 values) of (a, c, d, f) are the CASP influent values, while b and e refer to the PAC + UF influent.

While the ΔUV254 range achieved by the processes did not change by introducing PAC recirculation (no horizontal shift of ΔUV254 ranges in Fig. 5a–f), the OMP removal did increase in the CASP in the case of PAC recirculation, especially in the inline-dosing process (vertical data shift in Fig. 5a). At the same time, correlations within the PAC + UF stage remained very similar, regardless of recirculation (cf. Fig. 5b, e). Consequently, the overall inline-dosing process remained within the same ΔUV254 range, whereas OMP removal improves (cf. Fig. 5c). Interestingly, the Ulm process showed a similar behaviour in the CASP stage as the inline-dosing process, but with much smaller effect size (cf. Fig. 5d–f), while overall OMP removal was not significantly affected in this process (cf. Fig. 2).

In terms of practical application, we conclude that existing correlations between OMP removal and ΔUV254 during the adsorption stage remain valuable for estimating removal in PAC recirculation setups. UV254 can thus be measured in the cleaner matrices before and after the PAC + UF stage (as opposed to in the WWTP influent). This allows for measuring in a matrix that causes less fouling and therefore leads to less drift and maintenance requirements of online probes31. When using UV254 data from experiments or secondary effluent measurements as a reference, an additional OMP removal percentage can be attributed to PAC recirculation. Hence, a bonus removal percentage can be added to the estimated OMP removal of the CASP. For this purpose, we propose an increase in the OMP removal performance of the CASP from a commonly variable 20–30% removal in the CASP, to a more stable 45–55% (lower boundary of confidence band, at 95% confidence level, cf. Fig. 5a) for CASP setups with recirculated PAC, i.e., an addition of ~25% points.

Strong correlations of OMP removal and ΔUV254 were reported, even for simultaneous dosing set-ups, where the overall influent serves as reference17. While such probe installations proposedly come along with higher maintenance requirements to avoid signal drift, the recirculation process scheme could potentially also be controlled by overall ΔUV254.

Substance selection matters when evaluating removal performance

The UWWTD defines a selection of OMPs that aims at a representative and robust performance evaluation of the OMP removal stage (minimum average removal of 80%). In this study, we make use of this selection of OMPs to compare OMP removal performance. However, the UWWTD leaves some degrees of freedom in selecting a subset from a list of 12 substances. It is difficult to predict or exactly quantify the adsorption behaviour of these substances in advance, as it depends not only on chemical characteristics (e.g., molecular weight, hydrophobicity, log D, charge/ionisation) and concentration of a target substance, but also on its interactions with the adsorbent product governed by its properties (e.g., pore size distribution, point of zero charge, surface chemistry and heteroatoms), the surrounding water matrix properties (e.g., pH, temperature, ionic strength, competing substances) and the exact composition of the matrix as a multi-solute system32,33,34. Therefore, we included two representative substance selections: “EU Best Mix” and “EU weighted” which are both compliant with the UWWTD requirements. Which exact substance selection is to be used for monitoring and surveillance at a specific site is to be negotiated between the plant operator and the water authority. Relevant aspects in this negotiation comprise of abundance of individual substances (that must be high enough above the analytical limit of quantification to allow for robust process evaluation), baseline chemical state of the receiving waterbody, and site-specific priority substances (e.g. diclofenac to meet absolute environmental quality standards (EQS) in the receiving waterbody), amongst others and besides the physicochemical properties of the selected substances. On the one hand, the “EU Best Mix” selection (Fig. 2a–c) contains mostly well adsorbing substances and hence resulted in comparably low PAC dosages and thus most cost-effective operation to comply with UWWTD requirements. On the other hand, the “EU weighted” (Fig. 2d–f) selection contains a larger number of substances and substances with a broader spectrum of properties. This also includes not well adsorbing substances34 such as candesartan and irbesartan that both contribute to the average removal with relatively low individual removals (cf. Formula 1). Consequently, the resulting PAC dosage requirements were consistently higher in this case irrespectively of whether PAC was recirculated or not (cf. Fig. 2 a vs. d, b vs. e, and c vs. f). Most notably, the conventional PAC both, with and without recirculation, did not reach the 80% removal target under the tested conditions at all (Fig. 2e). Moreover, it could not be proven that the larger number of substances in the “EU weighted” selection as compared to the ”EU Best Mix” selection (9 vs. 6 substances) resulted in a lower variation of the performance results and thus a potentially more robust evaluation. Finally, it has to be emphasised that care must be taken when comparing removal results from different studies that are based on different substance selections.

Practical implications of PAC recirculation for the CASP and the UF process

Another factor that might have delayed the practical implementation of membrane hybrid processes, and especially PAC recirculation, is the relatively little reported influence of recirculated PAC on the upstream CASP. In this study, the CASP was operated at a constant mixed liquid solids concentration of ~3.5 g/L, also when PAC was recirculated. PAC recirculation was active for a period of 14 months, including adaptation times of several weeks prior to sampling each setting. This was sufficient to reach a steady state of both PAC concentration in CASP and to allow for a biological process adaptation. Similar treatment performance compared with a conventional CASP was observed for standard water quality parameters (COD, BOD, TSS, NH4-N, and PO4-P; cf. Fig. S2 and Text S3). The sludge settling index was analysed and improved when 20 mg/L PAC was dosed (cf. SI Text S5 and Figs. S8 and S9). Lower PAC dosages did not result in a consistently improved sludge settling index. However, the sludge settling properties were subject to changes over time. Presumably, the sludge properties of the pilot-scale biological treatment were too easily affected by rapidly changing influent properties (e.g. frequent industrial discharges) and seasonal temperature fluctuations to yield consistent results for the sludge settling index. Moreover, we observed that PAC was incorporated into sludge flocs (cf. Figs. S4 and S5). Overall, these results on sludge properties are in good agreement with other studies describing that PAC dosage can lead to improved sludge properties, e.g., enhanced dewatering of excess sludge, improved sludge settling index, and improved calorific value during PAC-added sludge35,36,37,38. Nevertheless, these improved sludge properties should be further verified during full-scale operation to exclude the variability of pilot systems.

No negative impact on permeability or backwash efficiency of the membrane could be observed (permeability range was maintained compared to Zimmermann et al.2), however exact quantitative comparison to the operation without PAC recirculation was not possible due to i) chronologically consecutive settings and, hence, different prevalent water matrices and ii) a membrane impacted by long-term fouling where no adequate measures to precisely quantify potential irreversible fouling effects were taken.

Preliminary economic and operational feasibility and limitations of this work

We conclude that a PAC recirculation in a CASP and PAC + UF system does not negatively impact the overall process and may offer economic advantages in terms of sludge disposal. In practise, the PAC recirculation could be pumped to the CASP by existing backwash pumps, resulting in no additional machine requirements. Since downstream ultrafiltration commonly recirculates physical backwash retentate back to the biological stage, no additional pumping energy for PAC recirculation must be accounted for, compared to an ultrafiltration process without PAC.

While the counter-flow scheme showed to be an ideal setting for the usage of fine PAC products, reaching very low dosage requirements, fine PAC products are potentially more expensive due to the extended milling requirements24. It should be kept in mind that such a recirculation scheme should only be applied if excess sludge is not used as agricultural fertiliser but instead is disposed of or incinerated.

Significance and prospect

The findings of this paper support the following conclusions in reference to our initial research questions:

  • PAC inline-dosing membrane hybrid processes show an improved removal of OMPs when partially loaded PAC is recirculated to the CASP stage applying counter-current principle. The required PAC dosage to meet UWWTD 80% OMP removal goals is, hereby, reduced by up to 50%. The inline-dosing process with application of fine PAC and PAC recirculation achieved compliance with the UWWTD requirements at a PAC dosage as low as 0.7 mgPAC/mgDOC. In contrast, the Ulm process (reference) achieved this performance at 1.0 mgPAC/mgDOC in our study, which aligns with existing literature. This fact highlights the remarkable performance of membrane hybrid processes for advanced wastewater treatment in the context of current EU legislation. Furthermore, it underlines that both, process configuration (with or without PAC recirculation) and PAC material selection play a pivotal role to improve overall process performance.

  • Kinetic experiments show: Inline-dosing processes with fine PAC can reach 30 to 50% of its maximum loading within the contact time in the pipeline (depending on the exact hydraulic contact time); another 20–30% of the maximum loading may be achieved while PAC remains in the membrane cake layer (depending on filtration duration and PAC availability to adsorption) and the remaining 20–50% can be loaded in the CASP after recirculation. Therefore, an improved reduction of OMPs by PAC recirculation can mainly be localised in the upstream CASP. Extended contact times and the counter-current principle were concluded as the main reasons for the improved removal.

  • While PAC recirculation impacts OMP removal and its correlation with ΔUV254, the bulk DOC is not impacted. This also indicates that the DOC composition changed in the different treatment stages due to PAC recirculation.

  • From a practitioner’s point of view, we proposed a way to transfer previous lab- or pilot-scale results (without PAC recirculation) to real applications with PAC recirculation in a straightforward approach of a bonus removal percentage that can be added when PAC is recirculated. This provides an early-stage estimate of the anticipated OMP removal performance in cases where PAC recirculation cannot be piloted on the same scale as the CASP. Moreover, in this study we were able to maintain a stable operation of the PAC + UF stage throughout the 2-year piloting.

As an outlook, the acquired data may serve as a reference for modelling approaches that aim to describe the complex adsorption behaviour in recirculation systems. Further research may clarify the proposed synergistic effects of fine PAC recirculation beyond improved adsorption alone, thereby enabling lower, more cost-effective carbon use. When the interactions between PAC and the coagulant, pore blockage, and biological degradation of adsorbed compounds are better understood, processes may be optimised even further.

Methods

The methodology of this work is based on the methodology of Zimmermann et al.2, which should be referred to for a detailed description of the pilot plant and its operation, as well as details on the utilised operational chemicals and PAC. In the following, we summarise only the most important aspects and such methods that differ from those of Zimmermann et al.2.

Pilot plant set-up

The pilot unit featured two parallel treatment trains that each consisted of a CASP and a subsequent PAC + UF membrane hybrid process stage. The pilot unit was operated from April 2022 until March 2024. The CASP was fed with mechanically pre-treated wastewater from the 250.000 PE WWTP Neuss Süd, Germany, at a constant flow of 450 L/h. The CASP featured a 3.4 m3 denitrification tank and an aerated 3.4 m3 nitrification tank, which were operated at a mixed liquor solids concentration of 3.5 ± 1.2 g/L in the nitrification stage, resulting in an average sludge age of 7 days. Excess sludge was separated in a 2.2 m deep sedimentation tank, where 27 L/h were withdrawn and disposed of as excess sludge and 450 L/h were returned to the denitrification tank as a combined internal recirculation and recirculating sludge.

Treatment train 1 was designed in an inline-dosing scheme, with no separate contact reactor. Here, PAC and water were mixed by the membrane feed pump, and the contact time equalled the retention time in the pipework between the PAC dosage point and the membrane. This contact time could be altered between 30 and 60 seconds, depending on the desired setting. Treatment train 2 was designed as a modified Ulm process. This train included a contact reactor for mixing of PAC and secondary effluent, a lamella clarifier for PAC sedimentation, including an internal recirculation to the contact reactor for PAC concentration within the adsorption stage. This setup offered an alterable contact time of 30 or 60 mins.

Both treatment trains were equipped with two hollow-fibre UF modules, each 1.7 m long, with a total active membrane surface area of 3.2 m2. The membrane was a DuPont™ inge™ ultrafiltration fibre, made from modified polyether sulphone with a nominal pore size of 0.02 µm. The modules were operated in an inside-to-outside, back-end filtration mode. Therefore, every 40 minutes, a physical backwash with filtrate water flushed solids that collected in the membrane cake layer out of the module (for details of operation and cleaning procedures see Zimmermann et al.2).

The major difference to Zimmermann et al.2) is that within this study, the backwash retentate was collected in a mixed, intermediate collection tank that could collect the entire volume of one physical backwash (~10 litres). Within this backwash concentrate, PAC could be found as fragile agglomerates in the shape of the membrane fibre capillaries (cf. SI Fig. S4). The entire batch of collected concentrate was pumped back into the CASP stage with a separate membrane pump at flow velocities above 1 m/s to minimise settling of PAC in the pipes (see pink arrow in flow chart in Fig. 6 and Fig. S4 in the SI for an impression of the PAC recirculation). The PAC agglomerates mostly lost their integrity during transportation to and mixing in the CASP, with few agglomerates still recognisable in sludge flocs (cf. SI Fig. S5). In full-scale membrane plants, a direct transport of the concentrate could be realised by a sufficiently sized backwash pump, so that no intermediate collection and no extra pump would be required.

Fig. 6: Process flow chart of the pilot scale CASP and PAC + UF process trains.
Fig. 6: Process flow chart of the pilot scale CASP and PAC + UF process trains.
Full size image

Treatment train 1 featured an inline-dosing process design, with no separate contact reactor. Contact of PAC and water occurred in the pipework between the dosing point and the membrane module. The membrane feed pump ensures proper mixing of PAC and water. Treatment train 2 was designed as a modified Ulm process. This process design featured a separate contact reactor and lamella clarifier unit to increase PAC contact time. The treatment trains both included a recirculation line to pump backwash water with partially loaded PAC into the upstream CASP stage (pink arrow). SP sampling point, TMP measuring point of trans membrane pressure, WQ measuring location of online quality parameters.

The operation settings of the membrane process and adsorption stage were selected to match those of Zimmermann et al.2 to compare the effect of the PAC recirculation on adsorption performance. A detailed overview of the selected settings and a timeline of process performance data for this study are provided in the supplementary material (see Text S2, Fig. S1 and Table S1).

Materials and operational chemicals

The PAC products used within this study were commercially available. One was a Chemviron Carbon Pulsorb WP260 (here referred to as “conventional PAC”, Median diameter of 28 µm), the other a Pulsorb WP260 UF (here referred to as “fine PAC”, Median diameter of 5 µm). The utilised PAC products were from the same product line; however, they did not necessarily originate from the same production batch. Both treatment trains featured a coagulation step before the membrane filtration, utilising polyaluminium chloride dosed at 2 mgAl/L. The modified Ulm process featured an additional coagulation step within the contact reactor, using 4 mgFe/L ferric chloride. The membrane plant performed a chemically enhanced backwash every 24 hours, using subsequent soaking steps in sodium hydroxide and sulphuric acid (details see Zimmermann et al.2).

Sampling and analyses

All presented data is based upon 24 h mixed composite samples that were taken at the sampling points indicated in the flow chart (cf. Fig. 6). Samples were taken with no time offset between the sampling points. Sampling was done by programmable samplers into 10 L stainless steel containers in a cooling chamber, which was kept at 4 °C. Samples were analysed within 24 h after completion of the sampling time.

The analyses of chemical parameters presented in this work were conducted according to the protocols described in Zimmermann et al.2. OMP removal quantification followed DIN EN ISO 21676:2022‑01. This included solid phase extraction onto Oasis HLB cartridges, Waters Corporation, USA; high-pressure liquid chromatography (Thermo Fisher Vanquish): column Hypersil Gold aQ, oven temperature of 40°C, injection volume: 10 µL, Thermo Fisher Scientific Inc., USA; electrospray ionisation: positive and negative mode and high-resolution mass spectrometry: IDX Orbitrap, Thermo Fisher Scientific Inc., USA. A list of all determined analytes and their respective CAS numbers can be found in Table S3 of the SI, including further information on relevant parameters of the analysis, such as retention times, ionisation mode and fragment masses. Further details of the LC method are compiled in table S4.

UV254 data within this paper is based upon laboratory analyses according to DIN 38404-3 on a Macherey+Nagel UV Vis Spectrophotometer NANOCOLOR UV/VIS II; DOC was determined with a Shimadzu TOC-L following DIN EN 1484 (1997-08). Filtration for LC, DOC and UV254 analysis was done with 0.45 µm RC (regenerated cellulose) from MF-Millipore.

Data processing

Average OMP removal was calculated as described in the following:

Substance selection:

The substance selections to describe average removals (e.g., in Fig. 2) have been selected according to European legislation1, to offer a highly relevant selection for practical transfer of the results. In addition, for a more mechanistic understanding, Fig. 4 represents single substances with very different properties, in order to identify mechanistic trends and make results more transferable to other substances of similar adsorption properties, also in practical settings.

  • EU Best Mix: represents a selection of six substances from the UWWTD substance list in the required ratio of 1:2 between the two categories. The substances selected were those that showed the best adsorptive removal performance within this study, namely the Sum of 4- and 5-methylbenzotriazole, benzotriazole, carbamazepine, clarithromycin, hydrochlorothiazide and metoprolol.

  • EU weighted: To avoid a selection bias of substances with an affinity to the selected treatment technology, when fulfilling the EU UWWTD substance selection, some Swiss cantons proposed an approach that includes all substances but weighs the average removals in a ratio of 2:1 between categories 1 and 2 instead of selecting substances. This approach was based upon data from many different Swiss municipal WWTPs with years of experience in quaternary treatment in full scale39.

    The calculation is demonstrated by formula 1:

    $$\begin{array}{l}Overall\,avrg.\,removal {=}\frac{Avrg.\,removal\,of\,Category\,1\,substances{\times }2{+}Avrg.\,removal\,of\,Category\,2\,substances{\times }1}{3}\end{array}\,\,$$
    (1)

    Herein, all measurable substances of the EU list are taken into account, therefore no biased selection must be done. In this apparent study, the following substances of the EU list were measurable: sum of 4- and 5-methylbenzotriazole, benzotriazole, candesartan, irbesartan, carbamazepine, clarithromycin, diclofenac, hydrochlorothiazide and metoprolol.

Calculation rules applied, based on ref. 40:

  • When calculating the average removal of the selected substances, removal values for substances that were below 5·LOQ in the influent were dropped from the calculation.

  • Removal of substances with effluent concentrations determined as below the detection limit are calculated with 0.5·LOQ as their effluent concentration.

  • Removals of substances that showed negative removal values were omitted from the calculation.

The authors therefore tried to aim for a most realistic removal value as relevant to the existing legal framework, which is especially relevant to the implementation in practice.

Removal data from ref. 2 has been recalculated according to the substance selection described above.

Adsorption kinetics experiment

Our adsorption kinetics experiment was conducted in accordance with the recommendations of Böhler (2019) and Jekel (2018). While these recommendations were written for adsorption isotherm experiments, the major difference here is that we tested a single dosage at multiple time points, rather than varying the dosage, all after 24 or 48 hours. Otherwise, the setup was very similar to that in the aforementioned literature. We used horizontal shakers in a temperature-controlled laboratory at 21 °C.

The examined fine PAC and conventional PAC were not derived from the same batch as the PAC used for piloting; however, they shared the same product names. PAC was dried and weighed to prepare 1 g/L PAC dosing suspensions, which were stirred on a magnetic stirrer for 20 minutes prior to the experiment to ensure homogeneity. The wastewater was the secondary effluent of the pilot CASP and was sampled one day before the experiment. It was filtered through 0.45 µm membrane filters in a vacuum filtration set-up and cooled in a fridge overnight to ensure a stable sample. The water was allowed to reach room temperature before the experiment began. To ensure enough headspace for proper mixing in the shaker, we added only 500 mL of sample into 1000 mL glass bottles. At the start of the experiment, we added 20 mL of PAC suspension to the sample, resulting in a final concentration of 20 mgPAC/L.

10 mL samples were then taken after 2, 5, 10, 25, 60 and 120 minutes, as well as after 24 and 48 hours. Samples were quickly filtered through 0.45 µm syringe filters after extraction. UV254 was analysed in accordance with DIN 38404-3.