Introduction

Growing concerns about the global climate change crisis have motivated efforts to reduce greenhouse gases (GHGs) emissions and develop renewable energy resources1,2. There are several renewable energy sources that utilize water for the production of energy, including hydropower, geothermal, pressure-retarded osmosis (PRO), and reverse electrodialysis (RED)3. Although hydropower and geothermal energy offer low cost of electricity production, they also have the significant drawback in terms of high GHGs emissions4. For example, hydropower has an associated cost of 0.05 USD/kWh and a level of efficiency exceeding 90%. However, it is responsible for the emission of 41 gCO2/kWh. The cost of geothermal energy is 0.07 USD/kWh, with an efficiency range of 10–20%. It is associated with the emission of 170 gCO2/kWh5.

On the other hand, RED and PRO, which produce electric energy by utilizing the salinity gradient between seawater and freshwater, have no impact on GHGs emissions and are environmentally friendly6. Therefore, they are regarded as highly suitable from standpoint of the water-energy nexus7. Furthermore, RED is an attractive and promising technology in comparison to PRO, due to a number of advantages. Although PRO exhibits superior energy efficiency and power density (Supplementary Table 1)3,8, RED has lower membrane fouling and pretreatment costs and can be integrated into various industrial processes, such as seawater desalination and wastewater treatment facilities9. Furthermore, RED directly converts salinity gradient energy into electrical energy, thus circumventing intermediate conversion stages10. Recently, the global power output obtainable using RED was estimated to 625 TWh yr−1, which corresponds to 3% of global electricity consumption11.

RED is based on an electrochemical process using ion exchange membranes (IEMs), in which two solutions of different salinity are mixed together and the Gibbs free energy is directly converted into electricity12. Specifically, RED generates electrical energy through a stack in which anion and cation exchange membranes (AEMs and CEMs) are alternately arranged. Using porous spacers (woven or nonwoven) and gaskets, compartments are created between the AEMs and CEMs to allow the flow of feed solutions. The thicknesses of the gaskets and spacers determine the compartment width and maintain a certain distance between the AEMs and CEMs during operation. When feed solutions with different ion concentrations flow to either side of an AEM or CEM, a Donnan potential is formed on the membrane owing to the Donnan exclusion effect. When cell pairs are stacked, the Donnan potential is accumulated and can be used to produce electricity. Thus, the ionic current, resulting from the salinity gradient between membranes, is converted into an electric current at the electrode by the reaction of a redox couple at electrodes connected to an external load13,14.

In the electrochemical process driving RED systems, mass transfer in the bulk solution is important to enhance process performance and to control the kinetics. Although RED can be performed using a diverse range of feed solutions, most research has focused on seawater and river water owing to their abundance and energy potential. To improve the electromotive forces generated by Donnan exclusion between IEMs, various combinations of feed solutions with high salinity gradients, such as reverse osmosis (RO) brine and tap water, have been investigated, and a number of pilot-scale case studies have been reported15,16.

For long-term operation on the pilot scale as well as the lab scale, it is important to characterize foulants and fouling phenomenon. Although this issue has received less attention in RED than in other fields, the fouling behavior during RED operation has been investigated with different feed solutions, such as natural seawater, river water17, and waste sewage effluents18, and in systems with or without spacers19. Various foulants containing colloids17, multivalent ions20, and organic substances21 have been reported to cause a decrease in RED power output16,22,23. Further, the removal of such foulants has been attempted using methods such as feedwater reversal and air sparging22.

The realization of practical applications has been hindered by problems associated with the IEMs, including low power density, low energy efficiency, and membrane fouling. Thus, IEMs with improved electrochemical properties for stack performance24 and reduced membrane fouling25 have been developed for RED systems. Further, many researchers have focused on the development of IEMs with low ionic resistance and high permselectivity, which are the main factors determining the RED power output. Notably, pore-filling membranes fabricated by impregnating microporous substrates with polymeric (or monomeric) electrolytes can achieve impressive power densities of 2.4 W m−2, 26 because of the high density of ion-exchange electrolytes in the ultrathin substrates.

The Jeju Global Research Center (JGRC) of the Korea Institute of Energy Research (KIER) is located in the eastern region of Jeju Island, which belongs to the Korean Peninsula. Jeju Island features an abundance of underground seawater filtered by the volcanic bedrock called lava seawater27. Furthermore, a seawater RO desalination plant located near the JGRC uses lava seawater to produce commercial drinking water and discharges RO brine (up to ~2000 tons day−1). It is well known that the low energy density of RED can be overcome by using solutions with high concentrations of NaCl, such as RO brine. Through a KIER project, we have focused on the optimization of RED operation using site-specific feed solutions and the development of RED stacks with pore-filling membranes28,29,30,31, mainly using feed solutions such as natural (lava) seawater13, RO brine, and sewage effluents.

The initial aim of this study was to investigate changes in ultrathin IEMs (KIER membranes) and the stack performance of an RED system in environments where a high salinity difference was maintained. In contrast, many previous studies have focused on achieving high energy efficiency using RO brine15,18. Although calculations have been performed to evaluate how extreme concentration differences affect ultrathin membranes of 20 μm or less32, to the best of our knowledge, no experimental results during long-term continuous operation have been reported (Supplementary Table 2). During this work, we also discovered that the deterioration in RED performance could be attributed to water entrapment inside the AEM near the cathode. This phenomenon differs from common fouling mechanisms in environments with high salinity gradients. Continuous operation experiments were used to obtain insights into this novel fouling mechanism. Based on these results, we proposed the use of outer membranes for shielding from the electrode rinse solutions and the development of new IEM combinations for the stack assembly.

Results and discussion

RED stack performance during continuous operation

An RED stack consisting of ultrathin pore-filling membranes was continuously operated using real RO brine and natural underground water under constant voltage (CV) conditions. Figure 1(a) shows the open-circuit voltage (OCV) and stack voltage measured under CV operation over 10 days. The maximum RED power is achieved when the external resistance matches the stack resistance. The OCV was the highest on the first day (4.3 V) and then gradually decreased to 3.2 V. In an artificial NaCl solution with the same salinity ratio, the theoretical OCV value was calculated as 15.9 V, which is much larger than the measured value of 4.3 V. This can be explained by the uphill transport phenomenon, which is one of the effects of multivalent ions. The influence of multivalent ions results in a decrease in OCV and power density due to uphill transport, meaning ion transport occurs against the concentration gradient33. The electrochemical potential of Mg²⁺ is low at low concentrations due to its divalency, whereas for Na⁺, it is the opposite. The high transfer rate of Na⁺ disrupts electroneutrality near the CEM, prompting an uphill transport of Mg²⁺ to restore electroneutrality33,34. As shown in Table 1, the RO brine included multivalent ions (Mg2+, Ca2+, and SO42−), whereas only small amounts of Na+ and Cl− were detected in the underground water. Interestingly, Mg2+ is the most abundant cation in RO brine, with the exception of Na+. Moreover, the OCV value in the system with combined RO brine and underground water, which included less than 5% Mg2+ and SO42−, was significantly lower than those of previously reported systems18. This can be clearly understood by comparing the ion composition and concentrations of multivalent ions in the RO brine used in this study with those in the RO brine from previous study (Supplementary Table 3). Given the relatively high concentration of multivalent ions present in the feedwater utilized in this study, it can be postulated that the markedly reduced OCV is a consequence of their induced uphill transport. The transport of ions is discussed in further detail in the next section.

Fig. 1: RED stack performance.
figure 1

a OCV measured before operation and stack voltage measured under CV conditions, (b) net power and net energy efficiency, and (c) stack resistance and current as a function of time during continuous operation.

Full size image
Table 1 Characteristics of the RO brine and underground water used in this study
Full size table

Figure 1(b) shows the net power output over 10 days. On the first day, the maximum net power (1.84 W) was observed, and the net power density was calculated to be 0.92 W m−2total membrane (1.84 W m−2cell pair), which is significantly higher than previously reported values18. This can be attributed to the utilization of KIER membranes, which exhibit significantly lower membrane resistance, a thin membrane thickness, and comparable permselectivity, as detailed in Supplementary Table 4. The KIER membrane demonstrated superior performance when tested with natural seawater, exhibiting higher power density and energy efficiency compared to thicker commercial membranes (Supplementary Note 1 and 2). These findings also indicate that ultrathin membranes may confer a performance advantage for RED over relatively short periods. Nevertheless, the subsequent gradual decrease in power output and the occurrence of water trapping give rise to questions regarding the viability of using thin membranes for long-term operation. The power output decreased gradually from the first to the fourth day and then remained nearly constant in the range of 0.7–0.9 W. As the net energy efficiency depends on the net power output (Eq. (7)), the same decreasing trend was observed for the net power output. After stabilization, the efficiency was 17.7%, which was more than 56% smaller than the initial efficiency (max. 40.8%). It can be explained by the effect of multivalent ions. As stated by Guo et al.34, the ions present in the influent water, including K⁺, Mg²⁺, Ca²⁺, and SO₄²⁻, have a deleterious effect on OCV and power density, with Ca²⁺, Mg²⁺, and SO₄²⁻ being the most significant in that order. Furthermore, a report from Avci et al.35 indicates that the presence of MgCl₂ at a concentration of 10% in the influent water results in a reduction of up to 20% in OCV and up to 60% in power density.

Figure 1(c) shows the stack resistance and current measured over 10 days. Under CV conditions, the generated current decreased as the stack resistance increased, which contributed to the decrease in the net power output (Fig. 1(b)). The stack resistance increased until the fourth day of operation, reaching 6.83 Ω, and then remained nearly constant. This is significant because Mg2+ concentrations negatively affect the performance of RED. The increase in membrane resistance of CEM is due to the electrostatic interactions between the anionic functional groups in the CEM and the Mg2+ ions35.

Ion transport through ultrathin pore-filling membranes for continuous RED operation

Figure 2(a) shows the conductivity of high concentration (HC) and low concentration (LC) feed solutions measured at the inlet and outlet. Based on this data, the NaCl concentrations of the solutions collected at the inlet and outlet of each HC and LC compartment were calculated, and the concentration difference between the inlet and outlet is shown in Fig. 2(b). For simplicity of calculation, the measured conductivity was assumed to originate from NaCl only. For the HC compartment, the NaCl concentration at the outlet was 16.4–14.0 g L−1 lower than that at the inlet. In contrast, for the LC compartment, the NaCl concentration at the outlet was 20.7–29.5 g L−1 higher than that at the inlet. Specifically, on the first day, a decrease in concentration of 16.4 g L−1 was observed between the inlet and outlet in the HC compartment, whereas an increase of 20.7 g L−1 was observed in the LC compartment. After 4 days, the concentration differences between HC and LC compartments became larger. The changes in NaCl concentration between the inlet and outlet can be explained by ion transport and osmotic water permeation36. If the permselectivities of the IEMs were 100%, the change in the NaCl concentration of the LC compartment effluent would correspond to that of the HC compartment effluent. However, as shown in Fig. 2(b), the change in concentration in the LC compartment was roughly twice that in the HC compartment. On the 4th day, the flow rate of the LC effluent was 15 mL min−1 lower than that of the influent, whereas an increase of ~5 mL min−1 was observed for the HC effluent. This observation indicates that co-ion and counterions penetrated the IEMs simultaneously. Furthermore, the mismatch between the volume decrease in the LC compartment and the volume increase in the HC compartment was likely due to the differences in loss to the electrode rinse solution (ERS). Unfortunately, the increase in the volume of the ERS could not be measured. Nevertheless, owing to this increase in volume, fresh ERS was used for each day of RED operation.

Fig. 2: Ionic transport during continuous operations of an RED stack.
figure 2

a Conductivity, (b) difference in concentration and volume between the inlet and outlet of HC and LC compartments under CV conditions, and (c) total mass transport of NaCl and coulombic contribution as functions of time. HC and LC correspond to RO brine and underground water, respectively.

Full size image

Under a high salinity gradient, osmotic water permeation from the LC compartment to the HC compartment could also contribute to changes in concentration. The ionic concentrations of the influents and effluents are summarized in Table 2. The concentrations of Na+ and Cl− in the HC influent did not correspond to those in the LC effluent. Thus, the concentrations of Na+ and Cl− in the LC effluent were higher than expected based on the amounts of these ions that passed from the HC influent to LC effluent. This result can be explained by co-ions passing through the IEMs together32,37.

Table 2 Ion concentrations in the influents and effluents of the RED stack on day 5
Full size table

Figure 2(c) shows the total mass transport of NaCl and the coulombic contribution calculated using Eq. (9). The total mass transport increased from 4.54 mmol m−2 s−1 on the first day to 5.37 mmol m−2 s−1 on the fourth day and decreased gradually, reaching 4.95 mmol m−2 s−1 on day 10. In contrast, the amount of NaCl transported by the electric field decreased from 0.11 mmol m−2 s−1 on the first day to 0.10 mmol m−2 s−1 on the fourth day and then remained constant at 0.06 mmol m−2 s−1 until day 10. Thus, in the IEMs, co-ion transport was dominant over counterion transport. The coulombic contribution to NaCl transport calculated based on the current generated in the RED stack indicated that the permselectivities of both the CEMs and AEMs were severely damaged after 4 days. As demonstrated by Tedesco et al. 32, when the membrane thickness is less than 20 μm, the increased flux of co-ions enhances the diffusive flux through the membrane, thereby reducing the current density (and current efficiency). These findings were corroborated in the present study, which also demonstrated a decline in the measured current values (Fig. 1(c)) and confirmed the dominant role of co-ions transport through Table 2 and Fig. 2(c).

Membrane fouling

After continuous operation, the RED stack was disassembled to characterize the change in membrane conditions. Figures 3 and 4 show scanning electron microscopy (SEM) images and energy-dispersive X-ray spectroscopy (EDS) spectra of membranes from three locations: the shielding membrane in contact with the cathode, a membrane randomly from within the stack, and the shielding membrane in contact with the anode. In the case of CEMs, irrespective of the location in the RED stack, no serious fouling was observed by SEM analysis, with the original membrane condition apparently maintained. However, EDS elemental analysis revealed that small amounts of inorganics from the RO brine, such as Ca and Mg, were deposited on the CEM surface (0.1 and 0.4 atom%, respectively). In contrast, in the case of AEMs, the SEM images revealed serious fouling on both the cathode and anode shielding membranes. Further, the EDS results confirmed that the foulants accumulated on the AEM surface comprised a large amount of Fe (up to 18 atom%) originating from the ERS and inorganics, such as Ca and Mg, derived from the RO brine. Although the AEM had a positive surface charge owing to the presence of cationic head groups, the detection of Ca and Mg on the AEM surface suggests that the permselectivity of the AEM decreased, which would facilitate the migration of co-ions through the AEM.

Fig. 3: CEM fouling.
figure 3

a–d SEM images (scale bars: 10 µm) and (e–h) EDS spectra of a CEM before and after continuous operation for 10 days.

Full size image
Fig. 4: AEM fouling.
figure 4

a–d SEM images (scale bars: 10 µm) and (e–h) EDS spectra of an AEM before and after continuous operation for 10 days.

Full size image

To investigate the origin of the deterioration in RED performance, the electrochemical properties of the membranes were evaluated (Table 3). Surprisingly, for the AEMs, the permselectivity decreased dramatically from 92% (pristine) to 67.0%, regardless of the location in the RED stack. Generally, permselectivity is known to be more sensitive to lower electrolyte concentrations because charge screening in the adjacent bulk electrolyte effectively suppresses Donnan exclusion in IEMs38. This trend is more pronounced for AEMs than for CEMs. Hence, a dilute interface can strongly influence co-ion exclusion by the membrane, and the salt concentration adjacent to the interface has a particularly strong effect on permselectivity39. As mentioned above, the coulombic contribution to the amount of NaCl transported, which depends on the permselectivity, decreased until the fourth day of RED operation (Fig. 2(c)), and it was unclear which IEM had a greater effect on this behavior. However, the results in Table 3 suggest that the decrease in the permselectivity of AEMs exposed to highly concentrated brine promoted the transport of co-ions rather than counterions. This deterioration in AEM performance was likely the dominant factor in decreasing the performance of the entire RED system, even when a high salinity gradient was maintained between the AEM and CEM. Moreover, compared to the pristine membrane, the AEMs on the anode and cathode sides showed a significant increase in membrane resistance, likely due to poisoning by the ERS.

Table 3 Electrochemical properties of membranes after continuous operation for 10 days (values are means ± standard deviations, n = 3)
Full size table

Water-trapping phenomenon within the AEM

Interestingly, we observed water trapped inside the AEM near the shield side of the cathode, which has not been reported previously during RED operation. To confirm this phenomenon, the experiments were repeated twice under the same conditions. Figure 5 shows photographic images of an AEM located near the cathode after RED operation for 10 days using RO brine and underground water. The porous substrate used in this study consisted of high-density polyethylene produced using a biaxial stretching process40. Cross-sectional SEM imaging revealed that this substrate was laminated with a multilayered structure (Supplementary Fig. 4). Although the IEMs were as thin as 16 µm, water could become trapped inside if the multilayer structure of the porous substrate was broken. Moreover, although no leakage of the ion exchange resin from the porous substrate was observed, it is possible that structural changes or ion exchange resin leakage occurred, as the ion exchange capacity (IEC) and permselectivity decreased (Table 3).

Fig. 5: Water-trapping phenomenon.
figure 5

Photographic images of water trapped inside the AEM near the cathode. The results of repeated experiments are displayed, showing differences in the location at which water was trapped.

Full size image

Compared to other commercial IEMs, the pore-filling membranes used in this study have may advantageous properties, such as low electrical resistance, high permselectivity, and low water permeability (Table 3). However, it should be noted that (1) the water permeability of the AEM was significantly lower than that of the CEM, (2) both the CEMs and AEMs exposed to a high salinity gradient owing to the combination of RO brine and underground water, and (3) the shielding CEM on the cathode side in particular was in contact with a HC electrolyte.

To explain the water-trapping phenomena inside the AEM near at the cathode, we hypothesize that an imbalance occurs between the AEM microstructure, consisting of a porous substrate and ion exchange resins, and the transport of osmotic water occurs within the AEM. Furthermore, it is thought that the high concentration of cations moving through the shielding membrane may affect the single-electron transfer reaction of ferri-/ferrocyanide in the ERS at the cathode and mass transfer inside the RED stack. A plausible explanation for the water-trapping phenomenon is illustrated in Fig. 6 based on various factors in (a) the membrane and (b) the electrochemical reaction in the ERS at the cathode. As shown in Fig. 6(a), both the CEM and AEM are in direct contact with the HC compartment, and ultrathin IEMs are known to reduce energy efficiency and power output through co-ion transport and uncontrolled mixing by water transport32,41,42. The pore-filling IEMs used in this study are also quite vulnerable to these effects. Table 4 summarizes the ion composition of the water trapped inside the AEM. In addition to increases in the concentrations of the anions moving through the AEM, the concentrations of cations such as Na+, Ca2+, and Mg2+ also increased, confirming that they are captured within the AEM. Thus, these results clearly demonstrate that considerable co-ion mixing occurs through the AEM.

Fig. 6: Origin of increased membrane resistance.
figure 6

a Plausible explanation for the water-trapping phenomenon in the RED stack, and (b) influence of increased metal cation concentration on the ferri-/ferrocyanide redox reaction at the cathode.

Full size image
Table 4 Ion composition of a feed solution extracted from within the AEM
Full size table

Notably, water transport by osmosis was also activated because the AEM was directly in contact with HC and LC electrolytes. Thus, excessive osmotic water transport occurred through the AEM. However, it is believed that an imbalance occurred between water transport accompanied by co-ion transport and the osmotic water flux, resulting in the water-trapping phenomenon inside the AEM. For osmotic water migration into the HC compartment, water molecules inside and at the interface of the AEM should be freely transported through the formation of water channels. However, IEMs in contact with HC brine shrink through dehydration (deswelling) and consequently exhibit increased membrane resistance43,44, which may hinder the ability the membrane to allow water to flow in one direction.

It is also important to understand why this phenomenon only occurs in the AEM near the cathode. Figure 6(b) shows the possible processes occurring near the cathode and shielding CEM. A large amount of cations (both mono- and divalent ions) move toward the electrode solution through the shielding CEM on the cathode side in contact with the HC compartment, creating an environment in which the reduction reaction is activated. The rate of the ferri-/ferrocyanide redox reaction is known to depend on the cationic concentration of the electrolyte36. Furthermore, the solvation structure of the ferri-/ferrocyanide redox couple depends on the presence of structure-making or structure-breaking ions. Cations such as Li+ and tetrabutylammonium are structure-making ions that interact strongly with water molecules, resulting in slower electron transfer kinetics during electrochemical processes at the electrode45. However, structure-breaking cations such as Na+ and K+ have the opposite effect and thus accelerated electron transfer kinetics46. Therefore, activation of the electrode reaction on the cathode side likely resulted in a breakdown of the electroneutrality within the RED stack. Excessive mass transfer through the CEM and AEM could act as a driving force to balance the electroneutrality of the entire system. The high salt concentration gradient across the AEM causes excess water molecules to move toward the AEM according to the osmotic pressure. However, the AEM not only has lower water permeability than the CEM but the high saline concentration is expected to increase the membrane resistance. Although the calculated water flux indicates that osmotic water flux occurred from the LC compartment to the HC compartment, the amount of water that could penetrate the AEM was limited. Consequently, some water molecules cannot pass through the AEM and become trapped within the AEM, gradually accumulating around the cathode where the electrochemical reaction occurs.

Practical implications

We attempted to analyze the increase in the electrical resistance of the IEMs caused by water trapped inside the polymer matrix and determine whether this phenomenon is responsible for the deterioration in RED performance. This phenomenon differs significantly from common fouling mechanisms. Unfortunately, the plausible explanation presented herein is not sufficient to completely understand the water-trapping phenomenon. Further elucidation of this phenomenon will likely require a more detailed electrochemical analysis. However, based on the results of this study, we suggest that the combination of IEMs close to both the cathode and anode should be selected based on properties such as water permeability rather than general membrane properties such as low membrane resistance and high permselectivity. Moreover, it is essential to eliminate multivalent ions through pretreatment, as they exert a detrimental effect on the membranes. In particular, the ultra-thin ion-exchange membranes utilized in this study were observed to be significantly influenced by co-ion leakage and uphill transport caused by multivalent ions. While ultra-thin membranes contribute to high RED power output, the development of thicker membranes should be pursued to mitigate the negative impact of feedwater. Furthermore, in order to achieve stable long-term RED stack performance under high salinity gradients, we suggest that it is necessary to divide the IEMs within the stack into regions near the electrodes and other regions.

Methods

Materials

Sodium chloride (NaCl), hydrochloric acid (HCl), sodium hydroxide (NaOH), potassium ferrocyanide trihydrate, and potassium ferricyanide were purchased from Daejung Chemical & Metals Co. (South Korea). All chemicals were used without further purification.

Feed solutions

As a HC feed solution, RO brine was obtained from a commercial desalination plant on Jeju Island, South Korea. The average conductivity of the RO brine was 86.5 mS cm−1. As an LC feed solution, underground water was used, which had an average conductivity of 1.0 mS cm−1. The conductivity, temperature, and ion composition of the RO brine and underground water are summarized in Table 1.

RED stack configuration

Each experiment was performed using a cross-flow-type RED stack with 100 cell pairs (Fig. 7). Both the anode and cathode consisted of titanium welded mesh coated with platinum (Pt/Ti, thickness: 1 mm, Wesco Electrode, South Korea) with an area of 100 cm2 (10 cm × 10 cm). The electrodes were connected to a polyvinyl chloride endplate. An aqueous solution of 50 mM K4[Fe(CN)6] ∙ 3H2O and 50 mM K3[Fe(CN)6] was used as the ERS. The IEMs used in this study (K-CEM and K-AEM) were produced at JGRC using a pore-filling method via a roll-to-roll process (Supplementary Fig. 5). The detailed production method has been described elsewhere28. The main properties of these IEMs are summarized in Table 3. Each IEM had a thickness of 16 µm and the effective membrane area was 2 m2. In the assembled stack, an intermembrane distance of 0.1 mm was achieved by inserting a reinforced polytetrafluoroethylene gasket (thickness: 0.1 mm, Alphaflon, South Korea) and mesh-type spacers (thickness: 80 μm, DS mesh, South Korea) between the IEMs, which prevented direct contact between the membranes.

Fig. 7: RED operation.
figure 7

a Illustration of experimental layout and (b) photographic image of the RED system used in this study for continuous operation.

Full size image

RED stack operation

The RED stack was continuously operated for 4 h per day for 10 days. The flow rates of the HC and LC feed solutions were 100 mL min−1, and the flow rate of the ERS was 50 mL min−1. The flow rates of the HC and LC feed solutions were automatically measured using ultrasonic flow sensors (VN05R, Aichi Tokei Denki Co., Japan), and pressure transmitters (P203GF, Allsensor, South Korea) on the inlets were used to automatically measure the pressure drop. To minimize impurities in the feed solutions, the RO brine and underground water were filtered through a cartridge filter (pore size: 0.01 mm, Synopex, South Korea) and a 1/4” mesh filter (pore size: 5 µm) before being fed into the RED stack. The filters were changed weekly, and the ERS was replaced with fresh solution daily before beginning RED operation. Three diaphragm pumps (0.2–1.3 L min−1, SIMDOS NF 1.100 TT.18RC, Switzerland) were used to inject the feed solutions and ERS, and the effluents were discharged from the RED stack without reuse. The ERS was recirculated between the cathode and anode compartments. The electrical potential, current, and power were measured using a DC electronic loader (ESL-300Z, E.L.P. Tek Co., South Korea).

Electrochemical measurements for RED performance

To evaluate the maximum power of the RED stack during continuous operation, CV conditions were used. The OCV was determined based on the stabilized value during the 10 min before continuous RED operation. The stack voltage (Vstack) applied under CV conditions was chosen to be half of the OCV value according to a previous report9.

For RED, the gross power output, Pgross (W), was calculated using the following equation:

$${P}_{{gross}}=I\times {V}_{{stack}}$$
(1)

where I (A) is the current measured under CV conditions.

The electric energy consumed by feeding the HC and LC solutions was expressed as the pumping power (Ppump), and the net power output (Pnet) was obtained by subtracting the pumping power from the gross power output, as follows:

$${P}_{{pump},i}=\Delta {P}_{i}\times {Q}_{i}$$
(2)
$${P}_{{pump},{total}}={P}_{{pump},{HC}}+{P}_{{pump},{LC}}$$
(3)
$${P}_{{net}}={P}_{{gross}}-{P}_{{pump},{total}}$$
(4)

where ∆P is the pressure drop (Pa) in each feed solution compartment, Q is the flow rate of the feed solutions (m3 s−1), and subscripts HC and LC indicate the feed solutions.

To determine the net energy efficiency, we first calculated the Gibbs free energy (P∆Gmix) corresponding to the exergy, which is the amount of theoretically available energy that can be extracted from a system under equilibrium conditions.

$${P}_{\Delta {Gmix}}=2{RT}\left({{Q}_{{HC}}C}_{{HC}}{\mathrm{ln}}\frac{{C}_{{HC}}}{{C}_{{mix}}}+{Q}_{{LC}}{C}_{{LC}}\frac{{C}_{{LC}}}{{C}_{{mix}}}\right)$$
(5)
$${C}_{{mix}}=\frac{{{Q}_{{HC}}C}_{{HC}}+{{Q}_{{LC}}C}_{{LC}}}{{Q}_{{HC}}+{Q}_{{LC}}}$$
(6)

where Q is the flow rate of the feed solution (m3 s−1), C is the concentration of the feed solution (mol m−3), R is the universal gas constant (8.314 J mol−1 K−1), and T is the temperature of feed solutions (K). Subscripts HC and LC indicate the feed solutions, and the subscript mix indicates the equilibrium concentration after perfect mixing.

The net energy efficiency (\({\eta }_{{net}}\)) was obtained by dividing Pnet by P∆Gmix, as follows:

$${\eta }_{{net}}\left( \% \right)=\frac{{P}_{{net}}}{{P}_{\Delta {Gmix}}}\times 100$$
(7)

where Pnet is the net power output obtained from the stack and P∆Gmix is the Gibbs free energy.

Evaluation of ion transport through the RED stack during continuous operation

After 2 h of operation, the conductivities of the influents and effluents were manually measured over 2 min using a conductivity meter (SevenMulti, Mettler Toledo, Switzerland). The same samples were used to measure the flow rates (mL min−1) of the effluents.

The total mass flux of NaCl transported from HC to LC compartments per area was calculated based on the conductivity of the LC effluent37.

$${T}_{{total}}=\left({\varPhi }_{{LC}}^{{out}}{C}_{{LC}}^{{out}}-{\varPhi }_{{LC}}^{{in}}{C}_{{LC}}^{{in}}\right)/{{{A}}}_{S}$$
(8)

where Ttotal is the total mass flux of NaCl (mol m−2s−1), Φ is the flow rate (m3 s−1), C is the concentration (mol m−3), and AS is the effective area (m2) of one cell pair. Subscripts in and out refer to the influent and effluent of the LC compartment, respectively.

Ttotal is the sum of the co-ion transport and coulombic components (Tcoul). Tcoul was calculated using following equation37:

$${T}_{{coul}}=\frac{I}{{{FA}}_{S}}$$
(9)

where F is the Faraday constant (96,485 C mol−1) and I is the current (A) measured for the RED stack.

Membrane characterization

Before characterization, the membranes were rinsed 3 times with deionized water and dried in an oven at 60 °C for 24 h. For membrane resistance measurements, the AEM or CEM was immersed in 0.5 M NaCl(aq) for 24 h. The sample was then inserted between the compartments of a clip cell, which consisted of a pair of flat Pt electrode (diameter: 10 mm). The resistances of the sample +0.5 M NaCl(aq) system (Rm) and 0.5 M NaCl(aq) (Rs) were measured by impedance spectroscopy with an LCR meter (DU-6011, Delta United Instruments, Taiwan) at a frequency of 1 kHz. Subsequently, the membrane resistance (R) was calculated as follows:

$$R=({R}_{m}-{R}_{s})\times {A}_{E}$$
(10)

where AE is the electrode area.

For permselectivity measurements, the samples were immersed in 0.017 M NaCl solution for 24 h and then placed in the middle of a two-compartment cell. The two compartments were filled in 0.5 and 0.017 M NaCl solutions. The membrane potential (Em) was measured using a multimeter connected to Ag/AgCl reference electrodes while stirring the solutions in both chambers. The transport number (t) of the counterions in the samples was calculated using Eq. (11), and then the permselectivity (α) was calculated using Eq. (12).

$$t=\frac{\left[{E}_{m}/\left(\frac{{RT}}{F}{\mathrm{ln}}\frac{{a}_{1}}{{a}_{2}}\right)\right]+1}{2}$$
(11)

where R is the gas constant (8.314 J mol−1 K−1), F is the Faraday constant (96,485 C mol−1), T is the absolute temperature (K), and a1 and a2 are the activities of the concentrated and dilute NaCl solutions (M), respectively.

$$\alpha =\frac{{t}_{M}^{m}-{t}_{M}^{s}}{{t}_{X}^{s}}\times 100 \%$$
(12)

where subscripts M and X refer to the counterions and co-ions in the membrane, respectively. Superscripts m and s refer to the membrane and solution phases, respectively.

For IEC measurements, the AEM or CEM sample was immersed in 50 mL of 2 N NaOH or 2 N HCl, respectively, to replace the Cl− ion of quaternary ammonium chloride with a hydroxide ion (OH−) or the sodium ion of sodium sulfonate with a hydrogen ion (H+). After 24 h, the samples were rinsed 5 times with deionized water and then immersed in 3 M NaCl for 24 h. The obtained solutions were titrated with 0.01 N HCl or 0.01 N NaOH using an autotitrator (848 Titrino Plus, Metrohm, Switzerland). The IEC (meq g−1) was calculated using Eq. (13):

$${\rm{IEC}}=(C\times V)/{M}_{{dry}}$$
(13)

where C is the concentration of HCl or NaOH (0.01 M), V is the volume of added HCl or NaOH at the equivalence point (mL), and Mdry is the weight of the dried membrane sample.

The pure water permeability was tested using a self-made cross-flow apparatus (effective membrane surface area: 18.75 cm2). Pure water was used as the feed solution at 30 bar for 30 min at 24 °C. The water permeability was calculated using Eq. (14):

$${\rm{Water}}\; {\rm{permeability}}=\frac{V}{A\Delta t\Delta P}$$
(14)

where V is the volume of permeated water (mL), A is the effective membrane surface area (m2), ∆t is the measurement time (h), and ∆P is the transmembrane pressure (bar).

Before and after continuous operation, the membrane surfaces were observed using SEM (MIRA3, TESCAN, Czechia). After continuous operation, the RED stack was disassembled to select membrane samples. The membranes were rinsed 3 times with deionized water, dried in an oven at 60 °C for 24 h, and coated with Pt using a sputter coater (Q150R S, Quorum Technologies, UK) for 120 s at 10 mA.