Reactive media constructed wetland for phosphorus removal: assessing the opportunity and challenges

Wastewater treatment in rural areas usually consists of passive or low-maintenance processes, such as trickling filters, rotating biological contactors and/or constructed wetlands (CW). However, these have limited ability to remove phosphorus (P) to the low levels (≤1 mg/L) that are required from the newly implemented water framework directive (WFD) (EU Water Framework Directive 2000) across Europe. Therefore, alternative options that can remove P whilst maintaining the low maintenance and chemical inputs in an economically viable way are highly desired. Adaptation of CW using reactive media to replace gravel has been investigated in this regard due to the media's ability to remove P from wastewater through precipitation and adsorption mechanisms (Vohla et al. 2011). A comparison of the media previously tested indicates that those rich in Ca/CaO, such as steel slags, show higher P retention capacities than the alternative media (Gubernat et al. 2020). Particular attention has been given to blast oxygen furnace (BOF) steel slags which are a waste material from steel making and hence offer a solution aligned to circular economic thinking.

Accordingly, there is a need to examine the performance of BOF slag, found to perform the best of all media tested, under realistic scales and conditions to establish the potential for the technology to be used at full scale and to validate the proposed mechanisms. The current paper aims to achieve this through the examination of a full-scale CW containing BOF steel slag (from a source not previously reported on) operated with a real municipal wastewater effluent providing more realistic treatment conditions of wastewater characteristics and P concentrations. The study lasted for 782 days and included analyses of the effluent quality, mineralogical properties of the media and reed development. To the authors' knowledge, this is the largest and longest reported study using BOF steel slag for P removal in CW.

A full-scale horizontal sub-surface flow wetland (Figure S1 in Supplementary Information) filled with BOF steel slag media (99,600 kg, d = 8–14 mm, Lafarge Tarmac Trading Ltd, UK) (no pre-treatment of the media was carried out) was operated as a trial plant for 782 days at a sewage treatment works in Leicestershire, UK. The wetland was built according to a typical tertiary design in the UK with a surface area of 100 m² (width: 8 m, length: 12.5 m) and 0.6 m depth. The bed was planted with Phragmites australis at 4 plants/m² which were not harvested throughout the trial. The influent of the system was the effluent of secondary clarifiers and P was supplemented to an influent P concentration of 5.8–9.5 mg P/L (Table 1). This would simulate the P concentration in a small sewage treatment works with no or limited P removal where a wetland would normally be installed. The bed was operated at a flow rate of 0.35–0.7 L/s equating to an EBCT of 24–48 h. The effluent from the system was blended with the effluent from other phosphorus removal trial plants and the effluent from the sewage treatment works prior to discharge to the receiving water to ensure that the regulatory requirements were met at all times at the site.

Online monitoring was set at the wetland influent and effluent for ortho-P, Fe³⁺, pH, turbidity, flow rate and temperature (influent only). Additional weekly grab samples were taken for analysis of P and chemical oxygen demand (COD) fractions (solid, colloidal, dissolved), total P (TP), total suspended solids (TSS), alkalinity pH, and metals. The metals analysed were Ca, Fe, V, Ni, Cu, Zn, As, Ag and Cd. The online analysis of P and Fe³⁺ was conducted according to colorimetric techniques at 15 min intervals (ABB online analysers, UK). Turbidity and pH were analysed with online sampling probes at 1-min intervals (Hach, UK). The additional samples were analysed through the standard methods with cell tests according to colorimetric methods for P, COD, Fe³⁺ and NH₄⁺ (Hach, UK). It should be noted that the samples were acidified for the Fe³⁺ and metals analysis to ensure measurement of the soluble forms. Repetition of some of the analyses with the online monitoring and grab samples allowed us to validate the results obtained. Fractions for P and COD were divided into unfiltered and filtrered through 1.2 μm and 10 kDa representing solid, colloidal and dissolved fractions, respectively. Suspended solids were measured according to standard methods. pH was measured with a handheld probe meter (VWR, UK). Metals were analysed with an ICP-MS (Perkin-Elmer, UK). All analytical tools were calibrated regularly and wherever applicable (e.g. ICP for the metals) certified standards were used and all samples were anlysed in triplicate to ensure accuracy of the measures.

Photographs of the wetland were taken weekly to monitor visual changes in plant growth over time and with the seasons. Rainfall and ambient temperature were monitored on site with a weather station.

Sequential P extraction experiments were carried out with steel slag samples following the method described by Letshwenyo (2014) using 1 M NH₄Cl, 0.1 M NaOH, 0.5 M NaHCO₃, 1 M HCl and concentrated HCl to desorb loosely bound P, Al bound P, Fe bound P, Ca bound P and P in stable residual pools, respectively. Samples of fresh, unused slag and exhausted slag at the end of the trial were analysed. Fresh steel slag samples were divided into unused (FU) and washed (FW), i.e. washed three times with 25 mL de-ionised (DI) water and air dried for 3 days. The exhausted slag samples were taken from four points in the direction of flow (Figure 2). All slag samples (1 g each) were air dried for 24 h before the commencement of the procedure, shaken for 24 h in each extraction solution and washed with 25 mL of supersaturated NaCl in-between steps. Analysis was conducted with respect to 0.45 μm filtered TP measurements. Solutions were also analysed for their metal content and the sum from each extraction solution was taken as the value of total extracted metal.

SI values were calculated in relation to the differences observed between inlet and outlet levels of calcium, ortho-P, magnesium, vanadium, aluminium, iron, silica, titanium, alkalinity and NH₄-N in combination with water temperature and pH on trial days 33, 121, 210, 253, 266, 329, 342, 352, 366, 406, 504, 581, 608, 685 and 749. Values of SI greater than 1 signified the potential to precipitate.

To complement this, the morphological and elemental composition of samples of unused and used steel slag was investigated using a scanning electron microscope coupled with an energy-dispersive spectrometer (ESEM, FEI XL30, Philips UK) operated with Aztec software (Oxford instruments NTS, Abingdon, UK). The scanning electron microscope (SEM) was operated at a voltage of 20 kv, spot 5 for imaging and analysis. The energy dispersive X-ray spectrometry (EDS), coupled to the SEM, was used to identify the elemental composition of the samples and provide information on their quantitative composition. The elemental composition was determined over the whole surface point of the sample. The EDS capture time was 60 s. Identification of the mineralogical composition of the media was conducted using an X-Ray diffractometer (Siemens, D5005 X-Ray diffractometer, UK). The XRD was operated at 10°–90° with slits of 2, 2 and 1 mm for an hour.

In phase 1 (days 1–209, P load from 0 to 0.47 mg P/g slag), the system achieved almost complete P removal with an effluent P concentration around 0.05 mg P/L (the detection limit). This coincided with a stable elevated pH of between 11 and 12 and a net calcium release of up to 37 mg/L (Figure 3(a) and 3(b)). Such conditions are consistent with the rapid dissolution of calcium oxide, leading to the formation of hydroxide ions and hence the increase in pH and subsequently precipitation of the released calcium (Johansson & Gustafsson 2000; Song et al. 2002). Towards the end of this phase of operation, the net calcium levels became negative indicative of continued precipitation and growth exceeding the rate of calcium release from the media and so utilising calcium from the incoming wastewater. In addition, other compounds were observed to be released from the media (Figure 3(b)). Most important is the metal vanadium which resulted in effluent concentrations of between 0.35 and 0.7 mg/L compared to an influent level that remained below 1 μg/L throughout. Vanadium levels in treated wastewater are not routinely consented as it is not expected to be present. However, potential limiting values around 60 μg/L have been discussed during other reported trials (Fonseca 2017). Accordingly, vanadium leaching poses a non-compliance concern if being considered for operational use. During this phase, substantial inhibition of reed development was observed due to competition with a weed identified as Epilobium hirsutum (Figure S2 in Supplementary Information). The plant is known to grow preferably in alkaline conditions whereas the reeds seemed dormant until the pH reached an acceptable level below 9 (Figure 3(a)) (Al-Farraj et al. 1984; Pérez-Fernández et al. 2006; Yin et al. 2016). At the back end of the bed, no reed growth was observed. It should be noted that according to the literature (Maucieri et al. 2020; Carrillo et al. 2022), plants have only a very limited contribution to phosphorus removal in CW and hence, it is assumed here that the removal observed throughout the trial was due to other mechanisms related to the presence of the reactive media.

In phase 3 (days 401–600, P load between 1.01 and 1.44 mg P/g slag), the performance was relatively stable with the pH remaining between 8.7 and 9.3 and the effluent P slowly decreasing from 2.76 to 1.24 mg P/L. This change coincided with an increase in the water temperature from 11 to 20 °C (Figure 3(c)). Net releases of calcium and vanadium were consistently below 23 and 0.1 mg/L, respectively, suggesting a slow but steady release from the media (Figure 3(b)). This indicates a stable period where reasonable P removal can occur, which is thought to be mainly associated with growth or adsorption onto the precipitate that had previously formed. During this phase, plant growth of reeds increased at the back of the bed and weed growth at the front and middle ceased (Figure S2 in Supplementary Information).

In phase 4 (days 601–782, P load between 1.44 and 1.86 mg P/g slag), the pH was stable with values between 8.9 and 8.7 and both the calcium and vanadium concentrations remained unchanged compared to phase 3. In contrast, a significant difference was observed for the effluent P concentration which increased up to a maximum of 7.33 mg P/L after 721 days of operation which meant nearly no P removal. From that point on, the effluent P concentration decreased rapidly again to reach a value of about 5.04–5.43 mg P/L at the end of the trial (782 days of operation, P load of 1.86 mg P/g slag) (Figure 3(a)). It should be noted that the influent P concentration was found to fluctuate significantly over this last period with spikes of up to 12 mg P/L. Fractionation of the effluent P revealed that the majority existed as dissolved phosphate denoting that it is unreacted or had desorbed (Figure 3(c)). As in phase 3, the changes appeared to coincide with temperature changes, which is congruent with previous reports of seasonal impacts (Figure 4(b)) (Shilton et al. 2006; Barca et al. 2013; Herrmann et al. 2014). The respective authors have associated the changes to either temperature-dependent P removal mechanisms, due to the increasing growth of algae raising the pH (Shilton et al. 2006) or the change in the solubility of calcium phosphate with temperature (Barca et al. 2013; Herrmann et al. 2014). This is supported by the fact that the reaction of calcium and phosphate is endothermic which means that the chemical equilibrium would be shifted towards the product side (calcium phosphate precipitates) at higher temperatures and result in higher P removal (Stumm & Morgan 1996). During this phase of operation, most plant growth ceased at the front of the bed and the reeds were established from the middle to the back of the bed (Figure S2 in Supplementary Information).

The reactive media bed was observed to also impact on other metals in the wastewater with iron, zinc, nickel, copper, arsenic, silver, cadmium and lead levels reduced by up to 84, 35, 18, 22, 41, 71, 52 and 15%, respectively. This occurred despite their low initial concentration (on average <0.2 mg Fe/L and <0.027 mg/L of all other metals) and indicates that other precipitates may have been formed other than calcium phosphate. However, the profile fluctuated through the trial relative to the different phases with uptake in phase 1 and release in phase 2 (Figure S3 in Supplementary Information). To illustrate, an uptake of zinc up to 56.9 μg/L was recorded in phases 1 and 3 while a release up to 35.7 μg/L was observed in phase 2. Uptake of other metals in phase one include 21.7 μg/L for copper, 1.8 μg/L for cadmium and 0.8 μg/L for lead. Comparatively lower release levels were observed in phase 2 at concentrations of 3.8, 0.32 and 0.51 μg/L, respectively. For arsenic, most variations were observed in phase 4 with fluctuations between 40.7 μg/L uptake and 40.9 μg/L release.

Further analysis of the link between pH and effluent phosphorus concentration revealed a decreasing trend (Figure 4(a)). Between pH 10 and 12, effluent P was stable below 0.25 mg/L and with decreasing pH from 10 to 8.3, the effluent P increased exponentially. Overall, it suggests that an effluent phosphate concentration below 2 mg/L only occurred when the pH was above 9. Below this level, the effluent P concentration changed significantly. In contrast, Park et al. (2017b) reported an ideal pH of >8 for P removal above 80% using steel slag which is consistent with the reported optimum pH range of 8.0–9.5 for calcium phosphate precipitation when the P concentration is below 5 mg P/L (Kim et al. 2006). The difference in the current dataset suggests a change in the removal pathway and a possible P removal mechanism related to the adsorption of phosphate to other precipitates (Giannimaras & Koutsoukos 1987). Further, if all removed P is assumed to have reacted to Ca₅(PO₄)₃OH, a stable pH between 10.1 and 11.1 can be predicted for the whole operational time. This only coincides with the actual measured pH in phase 1 suggesting that this might be the main P removal mechanism in that phase but not the subsequent ones.

Importantly, much of the previous work has been conducted in synthetic solutions and in the absence of carbonates. In the current case, the wastewater had an alkalinity of 178.1 ± 16.3 mg/L suggesting a relatively higher abundance of carbonates over phosphate. Accordingly, it is suggested that significant proportions of the precipitate were likely to be calcium carbonates. This was confirmed through SEM/EDX analysis of precipitates formed from a wash-out during commissioning, when no P was supplemented to the influent (data not shown) and more importantly during the analysis of the media taken from the bed (see next section on SI and materials analysis). In fact, Song et al. (2008) demonstrated an adverse impact of carbonate ions on calcium phosphate precipitation at pH 8 which, however, became negligible at more alkaline conditions (pH ≥ 9) showing that the pH is a key contributor. It is likely that lower P removal was not only caused by the pH decrease but also by less Ca²⁺ availability. The latter may have favoured CaCO₃ over calcium phosphate precipitation. For instance, Liira & Kõiv (2009) suggested that a doubling of retention time in shale ash columns resulted in supersaturation with respect to calcium carbonate. Despite being different media, steel slag and shale ash are both calcium-rich materials and it can be expected that the comparably high EBCT of this wetland (48 h during most of the study) did have an impact on calcium carbonate precipitation. Finally, these results highlight the need to monitor and understand the impact of carbonates on P removal and reactive media performance.

Across the entire trial potential supersaturation of 27–34 potential minerals was identified (Table S2 in Supplementary information). Many were associated with ferric oxides or calcium phosphates such as haematite (α-Fe₂O₃), HAP (Ca₅(PO₄)₃OH), magnesioferrite (MgFe₂O₄), maghemite (γ-Fe₂O₃), goethite (α-FeO(OH)), lepidocrocite (g-FeO(OH)), ferrihydrite (Fe₂O₃·0.5 H₂O) and Ca₃PO₄ (beta) as well as calcite.

This study reports on the first long-term, full-scale BOF wetland operated with real wastewater in the UK. An ultimate retention capacity of 1.39 mg P/g media was observed with low effluent phosphorus concentrations achieved for around 1 year of operation. The high removal efficiencies were associated with elevated pH and concentrations of other trace metals inferring rapid dissolution from the media surface. The high pH then promoted the reactions for phosphorus removal through precipitation and adsorption on formed precipitate. As precipitates formed on the surface of the media, the dissolution of the calcium oxide into calcium and hydrodide ions in the water was reduced which led to a decline in the pH and calcium concentration in the water. Once this declined, removal also declined and became more responsive to temperature changes.

Ultimately, the experiences outlined in the current study identify some inconsistencies with previously reported studies and the associated mechanism of removal. Current predictions state that the main P removal mechanism is calcium phosphate precipitation with a potential phase transitioning from amorphous to crystalline at a later stage when P removal decreases. From the results in this study, it can be assumed that calcium phosphate is one of the P removal mechanisms, but it is suggested that calcium carbonate as well as magnesium and iron minerals can also precipitate, and these minerals may present adsorption sites for P. Accordingly, the work suggests key consideration needs to be given to the impact of carbonates and the role of calcite on the removal of P from real wastewater.

In addition, the work has highlighted a number of challenges that need to be resolved before steel slag can be used as a P removal media in CWs. These include elevated pH, release of vanadium and the ability to sustain effective treatment for longer periods. It is posited that management of all three issues coincides with the ability to control the dissolution and release of calcium oxides from the media surface. Accordingly, the future focus needs to explore methods to deliver controlled precipitation onto the surface that mimics the natural weathering process but enables a controlled level of the initial release to maintain a lon lasting controlled and steady removal of the phosphorus.

The authors gratefully acknowledge financial support from Severn Trent Water. In addition, some of the work reported here was part of the AquaNES project which has received funding from the European Union's Horizon 2020 research and innovation programme under grant agreement no. 689450.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.