Stored hydrolyzed urine (Table 1) collected from Urine Diverting Dry Toilet (UDDT) in the eThekwini Municipality in Durban, South Africa, was used in this study as the feedstock (ethical clearance: BREC/00002684/2021). Urine was also diluted to simulate typical flushing scenarios from urine diversion toilets where flush volumes range between 0.5 and 2 L (von Münch and Winker, 2011). Assuming a urine volume of ∼300 ml (Haylen et al., 1989), dilution factors of 3 and 6 were prepared accordingly using deionized water.

Ammonium bicarbonate DS were prepared for each test run using distilled water and analytical grade ammonium bicarbonate. Draw solution concentrations of 0.6, 1.0, 2.0, 2.2, and 2.5 M ammonium bicarbonate solution were used in this study.

Thin film composite (TFC) membranes (OsMem ^TM TFC-ES, Hydration Technology Innovations) comprised of a polyamide active layer placed on the top of a thick microporous polysulfone support layer were employed in this study, due to their reported higher water fluxes and lower reverse solute flux (Kedwell et al., 2018; Almoalimi and Liu, 2022). The membrane supports a maximum operating temperature of 71°C and a maximum transmembrane pressure of 70 kPa. The pH of the solutions should be in the range of 2–11 and this high pH tolerance justified use of the TFC membrane for stored urine. Membrane sheets were immersed in deionized water before experiments as recommended by the membrane manufacturer. The membrane was supported in a SEPA crossflow membrane cell which allowed for tangential flow of both solutions (draw and feed) across the membrane. The cell was characterized by an effective filtration area of 140 cm², and a channel width and depth of 95.25 mm and 1.91 mm, respectively.

The membranes were operated in the AL-FS (Active Layer facing Feed Solution) orientation as this configuration minimized reverse solute flux in comparison to AL-DS (Active Layer facing Draw Solution), according to a baseline test using a 0.36 M sodium chloride solution FS with a 2 M ammonium bicarbonate DS over a period of 4 h. The measurements indicated a reverse solute flux of 4 ± 0.3 and 7.1 ± 0.1 g of ammonium bicarbonate per L of permeated water across the membrane in the AL-FS and AL-DS mode, respectively.

Feed and DS (1 L starting volume) were circulated at a cross flow velocity of 0.05 m s⁻¹ (Figure 1). The solutions were maintained at 22°C ± 0.1°C using a Grant TC120 thermostatic bath and continuously mixed with magnetic stirrers whilst sealed to ensure no loss of NH₄-N as NH₃ gas. The pH and conductivity of the FS as well as the pH of the DS were continuously measured through a probe Hach MM150, whilst the mass of both the feed and DS were measured on a Kern precision balance connected to a computer to log the data every two hours. Samples (5 ml) were taken every two hours during 8 h runs for chemical analysis. All FO runs were carried out in triplicate with new membranes for each run to evaluate the reproducibility of the results.

The following parameters were tracked during the FO runs: sodium (Na⁺), calcium (Ca₂ ⁺), magnesium (Mg₂ ⁺), potassium (K⁺), chloride (Cl⁻), sulfates (SO₄ ^2-) total nitrogen (TN) and total phosphates (TP). Sodium and chloride concentrations allowed for the initial characterization of the membrane rejections during the baseline runs. Potassium, total phosphates and total nitrogen concentrations were monitored due to their significance as agricultural supplements, while calcium, magnesium and sulfates gave an indication of precipitation that would have occurred during hydrolysis of urine. The feed and draw solutions were analyzed by concentrations and volume over time to calculate a mass balance and evaluate membrane rejections. The ions concentration (Ca⁺, Na⁺, Mg²⁺, K⁺) were determined using an Agilent 4,200 Microwave-Plasma Atomic Emission Spectroscopy, whilst the total phosphates total nitrogen, ammonical nitrogen and sulfates were determined using a Merck Pharo 300 Spectroquant. Chloride analysis was carried out using a Sherwood Scientific 926 chloride analyzer.

Water flux was calculated from the mass change of the FS over time, considering the density of water and area of the membrane Eq. (3): J w = Δ W ρ   A   Δ t (3) Where J _w represents the water flux (L·m⁻²·h⁻¹), ΔW is the mass change in FS (kg), ρ is the water density (kg·L⁻¹), A is the membrane area (m²) and Δt is the time interval between measurements (h). Overall permeate recovery R to the DS was calculated as a volume percentage at the end of each run according to Eq (4): R = V p V f × 100 % (4) Where V _p is the permeate volume (L) and V _f is the initial FS volume (L). Solute rejections were determined through Eq (5) based on the chemical analysis of the samples taken every two hours from the feed and DS. R ( t i ) = [ 1 − C s ( t i ) C f ( t i ) ] (5) Where R(t _i ) represents solute rejection at the time t _i (-), C f ( t i ) is the solute concentration in the FS at t _i (mg·L⁻¹), and C _s(t) is the solute concentration in the permeate at t _i (mg·L⁻¹). The solute concentration in the permeate was calculated through a mass balance by Eq (6) (Abousnina and Nghiem, 2014): C s ( t i ) = C d s ( t i ) V d s ( t i ) − C d s ( t i − 1 ) V d s ( t i − 1 ) V w ( t i ) (6) Where C d s ( t i ) is the solute concentration in the DS at time t _i (mg·L⁻¹), C d s ( t i − 1 ) is the solute concentration in the DS at time t _i-1 (mg·L⁻¹), V w ( t i ) is the permeate volume of water to the DS at time t _i (L), V d s ( t i ) is the volume of DS at time t _i (L) and V d s ( t i − 1 ) is the volume of DS at time t _i-1 (L).

Osmotic pressure of the feed and DS were determined from the measured concentrations of ionic species using PHREEQC™ software, which is a program based on the calculation of equilibrium chemistry of aqueous solutions.

Comparison of the water flux through the virgin membrane, fouled membrane (after several FO runs with undiluted stored urine accumulating to 48 h and a processed flux of 108.6 L m⁻²) and cleaned membrane can be used to evaluate the fouling and flux recovery through cleaning. Two cleaning methods were evaluated to test the fouling reversibility of the membranes after a run: the circulation of deionized water at higher flow rates, and the utilization of an osmotic backwash.

The first method involved cleaning the used membrane by circulation of deionized water on both sides of the FO rig for 30 min. Osmotic backwashing was also conducted on fouled membranes, whereby ammonium bicarbonate was circulated on FS side and distilled water on the DS side. This configuration leads to the inverse of the osmotic pressure gradient across the membrane, leading to a water flow in opposite direction compared to the forward osmosis runs. This is expected to remove the foulant material deposited on the membrane surface. Osmotic backwashes were conducted with 1 and 2 M ammonium bicarbonate solutions for 30 and 60 min in each case (Kim et al., 2012). Prior to osmotic backwashing, all the feed and DS trapped within the system were drained to minimize the influence of residual salts on the osmotic pressure differential across the membrane. Care was taken to not disturb the membrane prior to the backwashing. On completion of the osmotic backwashing, the flux recovery was measured, to have an indication of the flux recovered and to compare with deionized water operation.

Virgin and cleaned membranes were observed using scanning electron microscopy (SEM) with a maximum resolution of 50 nm, to determine the deposition of foulants on the membrane after cleaning. This analysis was coupled with energy-dispersive X-ray spectroscopy (EDX) to perform chemical analyses on selected areas from the virgin and cleaned used membranes. Prior to the SEM observations, the membranes were dried at room temperature and then coated with a thin layer of platinum using a sputter coater (SC7620, Emitech, United Kingdom).

Feed and DS concentrations were investigated to identify conditions which improved water flux, for increased FS volume reduction and water recovery prospects. Ammonium bicarbonate DS ranging from 0.6 M to 2.5 M resulted in increasing water fluxes from 1.4 to 4.7 L m⁻² h⁻¹ (Figure 2), driven by a greater osmotic gradient between the feed and DS. It would be expected that this relationship increases linearly, however at 2 M the line gradient decreases from 2 to 0.7. This suggests that at DS concentration gradients greater than ∼2 M, the onset of internal dilutive concentration polarization (ICP) occurs, whereby water accumulates within the support layer of an asymmetric membrane which in turn reduces the draw solute concentration at the membrane (Johnson et al., 2018). Several authors have reported this phenomenon in literature (McCutcheon et al., 2006; Le and Nunes, 2016). Therefore, 2 M was considered as the optimum DS concentration to be taken forward for further assessment. This concentration is also similar to Liu et al. (2016) who reported a water flux of 5.2 L m⁻² h⁻¹ using 2 M sodium chloride compared to 4.6 L m⁻² h⁻¹ with ammonium bicarbonate in this study.

As the experiments were operated in batch mode using 1 L draw and FS, water transport into the DS impacted the temporal water flux, due to the decline in osmotic gradient over time. Over an 8-h run using 2 M ammonium bicarbonate and stored urine, the transmembrane osmotic pressure decreased from 51.41 to 9.61 bar, which corresponded to water fluxes of 4.6 to 2.1 L m⁻² h⁻¹ (Figure 3). This decrease in the osmotic pressure gradient followed a logarithmic pattern, illustrating the declining rate in driving force and consequent water transport which is to be expected as the solutions verge towards equilibrium. Increasing the initial DS volume would prolong the solutions reaching equilibrium, thereby maintaining a higher osmotic pressure gradient/water flux and concentrating the FS at a faster rate. This attenuation of water flux decline by mitigating DS dilution was demonstrated by Xue et al. (2015), using a 2:1 DS:FS volume ratio with seawater and wastewater, achieving a 93% water reduction and a 10 fold concentration factor.

The implications of urine FS dilution with flush water were assessed in Figure 4. Greater dilution factors (DF) decrease urine osmolality, leading to an increase in transmembrane osmotic pressure. This results in a greater water flux where a DF of 6 reached 12.8 L m⁻² h⁻¹ compared to 4.6 L m⁻² h⁻¹ of undiluted urine, at the start of the run. Temporally, the water flux declines faster with increasing DF factors (linear gradients of −0.91, −0.51 and −0.31 for DF 6, DF 3 and DF 1 respectively, Figure 4). These results suggest that flush water is advantageous for FO favoring water recovery, however limits volume reduction and nutrient concentration due to the increased operational times required (Volpin et al., 2019b). In addition, a higher amount of energy will be required in the regeneration step of the draw solution. Indeed, a higher volume of water on the draw solution side will need more thermal energy for its heating to the temperatures where the ammonium bicarbonate decomposes into carbon dioxide and ammonium (<60°C).

The impact of water flux on key nutrient rejection was assessed on undiluted urine to understand the potential of combined water and nutrient recovery (Figure 5). Within an 8-h operating period, the membrane demonstrated high rejections of total phosphates (TP, 84%–94%), total nitrogen (TN, 72%–85%) and chloride (Cl⁻, 70%–85%), followed by potassium (K⁺, 39%–71%) and sodium (Na⁺, 32%–68 %). It is also important to note that the less critical ions Ca²⁺ and SO₄ ²⁺ and CO₃ ^2- were completely rejected during the run. The observed high rejections for TP and Cl⁻ could be attributed to their ability to retard the reverse permeation of the negatively charged bicarbonate ions from the DS, resulting in electrostatic repulsion as also reported by Zhang et al. (2014), Liu et al. (2016), and Volpin et al. (2018). The PO₄ ³⁻ ion was particularly retained due to its larger size (molecular weight of 95 g mol-1 compared to 35 g mol-1 of Cl⁻) aiding to steric hindrance associated to reduced membrane permeability (Zhang et al., 2014).

The membrane exhibited lower rejections of Na⁺ and K⁺, compared to Cl⁻, TN, and TP. A possible explanation for the low rejections of Na⁺ and K⁺ might be the fact that the membrane might have assumed a more negative charge as the pH of the urine feed stock increased during the FO process. The increase in pH (pH 7.5–8.4) could be attributed to the back diffusion of the ammonium bicarbonate draw solutes. The baseline tests (0.36 M NaCl FS and 2 M NH₄HCO₃ DS) indicated an average reverse solute flux of 4 ± 0.3 g of ammonium bicarbonate per litre of permeated water. The assumed negative charge would have attracted more positive ions towards the membrane (higher partitioning) resulting in higher permeation rates towards the DS as also observed by Zhang et al. (2014). Potassium also has a high permeability due to a smaller hydrated radius than PO₄ ³⁻, which has prevented high rejection in literature (Zhang et al., 2014; Ray et al., 2020). For the same reason, the small solute size of sodium has proven to be problematic as a NaCl DS, resulting in large reverse solute fluxes (Johnson et al., 2018).

Total nitrogen exists predominantly as ammonium ions in the stored hydrolyzed urine as the mean pH of the hydrolysed urine was pH 8 (Table 1), leading to only 4.39% ammonia because of a pKa of 9.34 at 22°C (Emerson et al., 1975). Therefore, following the explanations for the observed high rejections for negative ions and low rejections for positive ions, coupled with being the smallest ion (molecular weight = 18 g mol⁻¹), we could expect the TN rejections to be low. However, the rejections are higher than that of K⁺ and Na⁺. This could be attributed to the reduced mass transfer gradient that exists between the feed and DS with respect to the ammonium. The DS (ammonium bicarbonate) contains more ammonium ions than the stored urine, resulting in a reduced mass transfer force for ammonium diffusion from the feed to the DS, hence the observed higher rejections compared to the other positively charged ions. This phenomenon explains why Na+, Cl⁻ and K⁺ rejections are higher in other studies using NaCl and KH₂PO₄ as respective DS (Ray et al., 2020). Back diffusion could also be contributing to the enhanced TN FS concentrations (4 ± 0.3 g of ammonium bicarbonate per litre of permeated water). Therefore, the use of ammonium bicarbonate as a DS is particularly advantageous as a method for enhanced TN nutrient recovery, where TN is usually problematic to retain (Patel et al., 2020).

Generally, the rejections of nutrients declined with the water flux which decreased over time. This can be accounted for by the classical solution-diffusion theory, whereby the solute concentration gradient governs mass transfer. In the case of AL-FS FO, an increased co-transport of water dilutes the concentration of nutrients at the membrane interface (active layer and porous support) which results in a higher nutrient rejection (Zhang et al., 2014). The Na⁺ and K⁺ rejections decline the fastest with decreasing the water flux, which could be linked to the effect of their positive charge enhancing solute permeability compared to the negatively charged solutes.

Two in situ cleaning methods were evaluated for cleaning membranes to restore water flux after the same membrane sample was exposed to several forward osmosis runs with undiluted stored urine accumulating to 48 h and a processed flux of 108.6 L m⁻²: deionized water circulation (Figure 6) and osmotic backwashing (Figure 7). The fluxes were determined over 2 h with deionized water and 2 M ammonium bicarbonate as the FS and DS respectively. After circulating deionized water for 30 min, a fouled membrane with a flux reduced to ∼83% could be recovered to ∼95% of the flux using a virgin membrane (Figure 6).

For the osmotic backwash cleaning method which is particularly effective in controlling particulate and organic fouling (Kim et al., 2012), the duration and DS concentration were evaluated. The flux recovery improved by increasing the osmotic gradient between the feed and DS (i.e., by increasing ammonium bicarbonate concentration), and by increasing the duration of the osmotic backwash (only at 1 M DS concentration). An increase in the osmotic gradient results in a greater reverse water flux that removes foulants from the membrane which achieved ∼98% flux recovery at 2 M for 30 min. However, an unexpected decline in flux recovery during the fourth osmotic backwash after increasing the backwashing time to 60 min and ammonium bicarbonate concentration to 2 M was observed. This limit was also documented by Daly and Semiao (2020), who reported adhesive forces acting between the fouling layer and DS at high concentrations.

The foulants which could not be removed from DI water circulation and osmotic backwashing cleaning methods were examined using SEM and EDX analysis (Figure 8). The virgin membrane (Figures 8A,B) exhibited peaks of carbon, sulphur, and oxygen, which are the constituent elements of the membrane. Carbon has the highest content as it forms the backbone of the membrane structure. The membrane is composed of a TFC polyamide on polysulfone with an embedded polyester screen, which explains the O and S peaks. Through SEM observations, it was noted that some particulates remained linked to the membrane surface after cleaning (Figure 8C). Achilli et al. (2010), emphasized that the probability of mineral scale occurring on the membrane surface is enhanced when the FS is concentrated above the solubility limits of various water-soluble minerals. The FS for this work contained scale precursors such as magnesium, calcium, sulfate and bicarbonate ions. EDX analysis of the fouled membrane (Figure 8D) exhibited prevalent peaks of calcium, suggesting irreversible fouling by the deposition of bicarbonate on the membrane. In addition, sulfate fouling could occur, as this compound was completely rejected by the membrane and therefore prone to exceeding the solubility limits, causing scaling. However, the impact of irreversible foulants on the membrane was negligible for flux reduction during this study (3%–5% irreversible decline in flux), illustrating the low fouling propensity of the FO membranes and the ease of in situ membrane cleaning, which can significantly reduce the operating costs of the process.