Extending the Recovery Ratio of Brackish Water Desalination to Zero Liquid Discharge (>95%) Through Combination of Nanofiltration, 2-Stage Reverse-Osmosis, Silica Precipitation, and Mechanical Vapor Recompression

Abstract

Extending the recovery ratio (RR) of brackish water reverse osmosis (RO) plants to zero liquid discharge (ZLD, i.e., ≥95%) is vital, particularly inland, where the cost of safe retentate disposal is substantial. Various suggestions appear in the literature; however, many of these are impractical in the real world. Often, the limiting parameter that determines the maximal recovery is the SiO₂ concentration that develops in the RO retentate and the need to further desalinate the high osmotic pressure retentates produced in the process. This work combines well-proven treatment schemes to attain RR ≥ 95% at a realistic cost. The raw brackish water undergoes first a 94% recovery nanofiltration (NF) step, whose permeate undergoes a further 88-RR RO step. To increase the overall RR, the retentate of the 1st RO step undergoes SiO₂ removal performed via iron electro-dissolution and then a 2nd, 43% recovery, RO pass. The retentate of this step is combined with the NF retentate, and the mix is treated with mechanical vapor recompression (MVR) (RR = 62.7%). The results show that >95% recovery can be attained by the suggested process at an overall cost of ~USD 0.70/m³. This is ~60% higher than the USD 0.44/m³ calculated for the baseline operation (RR = 82.7%), making the concept feasible when either the increase in the plant’s capacity is regulatorily requested, or when the available retentate discharge method is very costly. The cost assessment accuracy was approximated at >80%.

1. Introduction

The reverse osmosis desalination of brackish water (BWRO) has become an important potable water source, especially in areas located far from the sea [1]. Since in many cases the generated retentate cannot be readily disposed of, a need has arisen to maximize the recovery ratio of the process, to reduce retentate disposal costs [2]. Typical BWRO processes are limited to a recovery ratio (RR) of 75–80%, due to chemical fouling limitations stemming (chiefly) from the precipitation of gypsum, fluorite, baryte, and calcite. The relatively low osmotic pressure of most BW feeds allows for much higher RR, but the limit imposed by the potentially precipitating solids must first be overcome.

To this end, quite a few treatment schemes have been suggested in the literature, involving almost all the stages in the reverse osmosis (RO) process, i.e., treatment of the feed, brine(s), and combinations thereof. For example, [3,4,5] used a combination of a nanofiltration (NF) pre-treatment of the BW feed to reduce the concentration of silica and multivalent cations (often termed “total hardness” or TH) and sulphate ions, thereby enabling a higher RR for the RO step that is applied on the NF permeate. Forward osmosis (FO) and recycling of an NF brine have also been suggested to increase RR and to add minerals to the BWRO product [3,5]. The authors of [6,7] used cationic ion exchange (CIX) and chemical softening to reduce the TH concentration of the feed. Other approaches involved combinations of different desalination techniques. For example, [8,9,10] used a combination of RO steps in various ways to decrease the energy demand and increase the overall RR. The authors of [11] suggested an approach that is based on low salt rejection RO membranes (LSRRO) to attain zero liquid discharge (ZLD) in both SWRO and BWRO applications. A short summary of notable techniques that were suggested for increasing the RR of BWRO is presented in

Table 1. Summary of treatment schemes suggested for increasing the RR of BWRO.

Treatment Train	Description	Final RR
>NF-FO-BWRO [1]	>A hybrid system combining nanofiltration (NF), forward osmosis (FO), and brackish water reverse osmosis (BWRO). This system utilizes the high rejection rate of NF for salt removal and the efficiency of FO for water extraction, improving the overall recovery of the RO system.	>~90%
>HBSRO [8]	>A hybrid batch/semi-batch reverse osmosis (HBSRO) system for high-recovery desalination. It involves three phases—semi-batch pressurization, batch pressurization, and purge-and-refill—resulting in a compact system size and lower energy consumption at high recovery rates.	>>90%
>CIX + RO (with seawater regeneration of CIX resin) [6]	>This process involves a cation exchange (CIX) unit to remove calcium and magnesium ions from the brackish water before it enters the RO unit. The CIX resin is regenerated using seawater, reducing the need for fresh water and increasing cost-effectiveness. This approach can increase the recovery ratio from 78% to 89%.	>89%
>NF-RO + blending [15]	>This process utilizes NF to pre-treat the brackish water and remove a portion of the salts. The NF permeate is then blended with the RO permeate to adjust the final salinity and mineral content of the product water, while achieving a high overall recovery ratio.	>95%
>Two-stage RO [9]	>This process is based on two RO units in series. The first stage operates at a lower recovery ratio, producing a permeate stream and a concentrate stream. The concentrate stream from the first stage is then fed to the second stage, which operates at a higher recovery ratio, further increasing the overall recovery.	>80%
>Closed circuit desalination [10]	>This process involves recirculating the RO concentrate back to the feed water, minimizing the volume of brine that needs to be discharged. It often involves using additional treatment methods, such as evaporation or crystallization, to manage the increasing salinity in the recirculating stream.	>97%
>APS + secondary RO [7]	>This method utilizes accelerated precipitation softening (APS) to remove hardness ions from the brackish water before it enters the first RO unit. This pre-treatment reduces scaling potential and allows for a higher recovery ratio in the secondary RO stage.	>98%
>Intermediate chemical demineralization + RO [4]	>This process utilizes chemical treatment for partial demineralization of the brackish water, typically by removing hardness ions. The treated water then undergoes RO for further desalination, resulting in a higher overall water recovery. The process increases the recovery ratio from 85% to 95%.	>95%
>NF-RO and Mg mineralization [5]	>The study proposes a hybrid nanofiltration-reverse osmosis scheme to enhance the mineral composition of desalinated brackish water, demonstrating economic viability and recovery ratios as high as 85%.	>85%
LSRRO [11]	The study employs low salt rejection membranes, placed in series to desalinate brine into a highly saline water (up to 4 mol/L) in small intervals, using a normal RO pressure of 70 bars, thus enabling an energy-efficient (<4 kWh/m3) desalination.	>95% (ZLD)

. The problem with silica precipitation and scaling in BWRO and prevention, thereof, is summarized in numerous other papers (e.g., [12,13,14]).

Different types of evaporation units are commonly practiced worldwide. These methods include mechanical vapor recompression (MVR), which has been extensively investigated and included as a part of various desalination schemes, e.g., [16]. MVR serves in this paper as an example for an evaporation technique, and its choice does not imply that the authors think it is the only or best alternative under any circumstance. It is, however, believed that the cost difference between using MVR, as opposed to other evaporation processes, is low and would not change the fundamental conclusions arising from this work.

The current work focused on utilizing conventional RO and NF trains, along with other proven water treatment techniques, to increase the RR of typical BWRO plants to beyond 95%, which is defined as ZLD. While other suggested methods, that rely on membrane fabrication/modification, as well as other innovative ideas [e.g., LSRRO], may be highly promising but not yet commercially applicable, the current study focuses on presenting quantifying and cost-assessing a readily available, feasible, and economic approach, that would, very likely, work in practice at large scales. The proposed process scheme was tested empirically (apart from the MVR step that, for technical limitations, was merely simulated, based on applied engineering practice) using real brackish water, a pilot-scale apparatus, and commercial membranes, to strengthen the credibility of the results and assess the related costs.

The work was aimed at increasing the recovery ratio (RR) of a previously published NF-RO filtration scheme [5], which utilized both the intrinsic monovalent–divalent ion selectivity of NF membranes and the high water–ion selectivity of RO, to ZLD, i.e., RR > 95%, while studying the baseline and marginal CAPEX and OPEX leading to ZLD operation. The limiting factor in the previous setup (developed in our group) was the high precipitation potential of silica (SiO_2(s)), which dictated a maximal RR of ~86%. By removing SiO₂ from the RO retentate, followed by applying a second RO pass on this stream and MVR to the NF and RO retentates, we aimed at increasing the RR to >95% (ZLD), which is increasingly required for minimizing retentate treatment costs, particularly in inland BWRO.

In the process that was presented by us in [5], the feed stream to the RO step underwent first an NF step to reduce the sulfate ion concentration, thereby mitigating the risk of gypsum precipitation and scaling at the high RR applied in the RO step. The subsequent limiting factor became silica (SiO₂) precipitation, with a recommended maximum concentration of ~200 mg/L when tailor-made antiscalants are used [6]. Given that the initial silica concentration in the feed was ~20 mg/L (a typical value in many brackish waters) and considering the very low, down to almost nil, rejection of silica by typical NF membranes, a maximal RR of 90% could be set in the RO step, bringing the maximal overall RR of the process to ~86%.

The brackish water source in the study was the Ma’agan Michael (Israel) groundwater, characterized by total dissolved solids (TDS) of ~5000 mg/L (see

Table 2. Chemical composition (main components) of the brackish water in Ma’agan Michael (Kantor) desalination plant [5].

Parameter	Unit	Value
Ca2+	mg/L	201
Mg2+	mg/L	198
Na+	mg/L	1378
K+	mg/L	44
Ba2+	mg/L	0.11
Sr2+	mg/L	-
SO42−	mg/L	3852
Cl−	mg/L	1970
F−	mg/L	0.5
NO3−	mg/L	6
SiO2	mg/L	20
Alkalinity	mg/L as CaCO3	305
pH	(-)	6.90

for a full chemical water composition). Applying a ~90% recovery RO step (on the NF permeate) would bring the TDS of the retentate to ~28 g/L (osmotic pressure of ~25 bar), thus the maximal RR of a further RO stage applied on this retentate should be close to 50%, to not exceed the maximal SWRO membrane pressure (70 bar).

The method tested herein as a pre-treatment step on the RO retentate to reduce the silica concentration involved the co-precipitation of SiO₂ with an iron hydroxide (either Fe(II) or Fe(III) based), with electro-dissolution (termed ED herein) as the Fe dosage method and MVR treatment of the produced NF and RO retentates. The overall suggested process is presented in Figure 1.

The main purpose of the paper, apart from presenting a new combination of process steps, is to quantify, cost-wise, the various parts that may lead to a BWRO ZLD operation, while elucidating the real costs incurred when attempting to elevate the recovery ratio from the traditional 80% to close to 90% and then to >95%. To this end, we attempted a combined empirical/simulative approach that we believe resulted in a reasonably (>80%), accurate total cost estimation (CAPEX + OPEX) and, perhaps more importantly, a sensible trend that could be used by professionals and decision makers.

2. Materials and Methods

The evaluation of the proposed scheme was divided into the following five main parts: (1) experimental NF and RO filtration using real brackish water and then a 2nd RO pass on the 1st pass retentate from which SiO₂ was removed; (2) Fe electro-dissolution experiments to test and cost-assess the SiO₂ removal step, using synthetic and actual RO retentate streams; (3) computer simulations using the data from the first part to evaluate the performance of the RO steps, to characterize the SI value of the possible solids that can precipitate and foul the membrane, along with the osmotic pressure that would develop in the ultimate membrane; (4) simulation/pricing of the MVR step that was suggested to further desalinate the mixture of the NF and 2nd RO stage retentates; and (5) detailed cost analysis through CAPEX and OPEX segmentation.

2.1. Description of the NF and RO Apparatuses and of the Experimental Procedure

Real brackish water (the feed stream in Figure 1) was collected directly from the well that supplies the Kantor desalination plant in Ma’agan Michael, Israel, whose composition is shown in Table 2.

All the NF and RO experiments were carried out using a pilot-scale desalination unit sustaining one 4″ high-pressure vessel, an Osip riva-80 booster pump, and a Hydra-Cell G10 high-pressure pump. A 10-micron filter was used to protect the high-pressure pump. The temperature was maintained at 25 ± 1 °C by a titanium heat exchanger and a chiller.

For the BW pre-treatment step (step 1 in Figure 1), an NF membrane (GE DL-4040-F1021 Stinger,

Table 3. Properties of the DL NF and SW30HRLE RO membranes, as reported by the manufacturers and in the literature.

Parameter	Unit	DL	Reference	SW30HRLE	Reference
Molecular weight cut-off (MWCO)	Da	150–300	GE Osmonics	-
MgSO4 rejection	%	98	GE Osmonics	99.8	FilmTec
Polymer type	-	Polyamide (TFC)	GE Osmonics	Polyamide (TFC)	FilmTec
Water permeability	L m−2 h−1 bar−1	3.5–10	[17,18,19,20,21,22]	0.9–1	[23]
Na+ rejection	%	5–30	[19,24,25]	99.8	FilmTec
Mg2+ rejection	%	20–50	[24]	99.8	FilmTec
Zeta potential (at pH 7)	mV	~ −20	[26,27]	−15–−30	[23]
Effective pore radius	nm	0.58–0.7	[19,28]	-

) was selected to allow good separation between the Na⁺ and Cl⁻ and the divalent ions (Ca²⁺, Ba²⁺, Mg²⁺, and SO₄²⁻). An antiscalant (Genesys CAS) for minimizing the scaling of CaSO_4, BaCO₃, and CaCO₃ was added at ~1 mg/L to the feed, resulting in a concentration of ~15 mg/L in the retentate produced in the 2nd-pass RO.

For the RO desalination steps (steps 2 and 4 in Figure 1), a SWRO (FilmTec SW30HRLE-4040, see Table 3) membrane was selected, to allow for a good separation performance in both desalination steps and to sustain the high pressure in the 2nd RO pass. Vitec^TM 4000, a silica inhibitor antiscalant was added at an initial concentration of 3 mg/L in all the RO experiments (the dosage recommended by the manufacturer is 2–5 mg/L).

The recovery ratio (RR) was defined as follows:

$$RR=Q_{\mathit{Permeate}}/Q_{\mathit{Feed}}$$

(1)

where Q_Permeate and Q_Feed are the permeate and feed flow rates, respectively.

Samples of the accumulated brine and permeate were taken for elemental analysis at different recovery ratios, where each RR was determined using Equation (1) and via determining the initial and final feed volumes in the holding tank of the desalination system. The final RO operation on the EC-treated RO brine (step 4 in Figure 1) was repeated three times (n = 3). The concentrations of relevant cations and of S (i.e., SO₄²⁻) were measured using a PlasmaQuant PQ 9000 Elite, High-Resolution Array ICP-OES (Analytik Jena AG, Langewiesen, Germany). The chloride ion concentration was determined using the Argentometric method [29]. The pilot-scale system was operated in batch mode; i.e., the retentate was circulated back into the feed tank. The single membrane module was set for a recovery ratio of 10%, as recommended by the membrane manufacturer [30]. The required NF pressure to obtain the corresponding flux to this recovery ratio was determined empirically, and the maximal applied pressure was 8 bar.

2.2. Iron Electro-Dissolution Experiments

2.2.1. Experimental Setup

Two iron electrodes (dimensions 80 × 135 mm; thickness 7 mm) were cut into shape and polished to serve both as cathode and anode in the iron electro-dissolution (IED) reactor. The IED step was applied at batch mode, using the electrode setup inside a beaker.

2.2.2. Jar Test SiO₂ Removal Experiments

To establish the required range of iron dosage for the co-precipitation of silica, a jar test was carried out with various iron concentrations using a synthetic RO retentate solution.

The alkalinity of the solution was set to maintain a pH slightly higher than 6, optimal for ferrihydrite precipitation, while maintaining negative CaCO₃ precipitation potential.

Two types of iron dosage techniques were tested within the jar experiments—direct ferric chloride (FeCl₃) dosage and electro-dissolution (IED, i.e., anodic reaction resulting in metallic iron dissolution). During the IED, ferrous ions (Fe²⁺) are introduced to the solution from the anode dissolution process; however, at pH values higher than ~5 in the presence of dissolved oxygen, Fe²⁺ rapidly oxidize into ferric ions [31] and precipitate as amorphous ferric hydroxide (Fe(OH)_3(s)). Equations (2)–(6) describe the process equations for the redox reactions (Equations (2)–(5)) and ferric hydroxide precipitation (Equation (6)):

$$F{e}_s\leftrightarrow F{e}^{2+}+2e^{-}, E_0=0.44 V \mathrm{vs}. SHE$$

(2)

$$F{e}^{2+}\leftrightarrow F{e}^{3+}+e^{-}, E_0=0.77 V \mathrm{vs}. SHE$$

(3)

$$2H^{+}\leftrightarrow H_2, { E}_0=0 V \mathrm{vs}. SHE$$

(4)

$$O_2+4H^{+}+4e^{-}\leftrightarrow 2H_{2}O, E_0=1.23 V \mathrm{vs}. SHE$$

(5)

$$Fe{\left(OH\right)}_3\leftrightarrow F{e}^{3+}+3O{H}^{-}, { K}_{SP}=6.3·{10}^{-38}$$

(6)

Silica was added to the synthetic brine solution to result in a concentration of 150 mg/L. HCl was added to neutralize the metasilicate solution pH and prevent increased alkalinity due to the Na₂SiO₃ addition (Na₂SiO₃ + H₂O → 2Na⁺ + 2OH⁻ + SiO₂). Next, 15 mg/L of Vitec^TM 4000 AS were added to the synthetic brine solution, which was stirred and left to fully dissolve and stabilize for two hours before each experiment.

Two jar test experiments were performed—an initial test (10, 20, 40, 60, 90, and 120 mgFe/L), to compare the difference in silica removal between the two techniques and to assess the required iron dosage range, and a main test, with dosages of 80, 110, 140, and 170 mgFe/L, corresponding to Fe/SiO₂ molar ratios of 0.6, 0.8, 1, and 1.2, respectively.

The iron salt (FeCl₃) was dissolved to form a 1 M solution, which was then introduced to the test beakers at the required volume. A constant current was applied in the EC experiments for varying times, and the overall iron dosage was calculated by the following equation:

$$C={\displaystyle \frac{t·I·\lambda }{V·z·F}}$$

(7)

where C is the molar concentration of the iron dosage; V is the volume of the solution (L); z is the electron valency of the iron species (mol e⁻/mol Fe); F is the Faraday constant (C/mol e⁻); I is the applied electrical current (A; C/s); λ is the current efficiency with value ranging from 0 to 1; and t is the required time in seconds.

Oxygen was supplied using a diffuser connected to an air pump during the first five minutes of the EC step, to both enable rapid ferrous ion oxidation (Equation (3)) and perform a rapid mixing step.

The experimental procedures relating to the silica co-precipitation were as follows:

Chemical addition—solution volume of 400 mL, 2 min rapid mixing at 100 RPM, 20 min slow mixing at 10 RPM, 25 min settling time.

Fe electro-dissolution tests—solution volume of 600 mL, 5 min rapid mixing using air bubbles for allowing enough time for complete Fe²⁺ oxidation, 20 min slow mixing at 10 RPM, 25 min settling time, ambient temperature, assumed current efficiency—100%.

2.2.3. Fe Electro-Dissolution Treatment of the Real NF-BWRO Brine

The produced NF/BWRO retentate was treated in the electro-coagulation reactor. A 170 mgFe/L dosage was chosen as the safety factor to ensure minimal silica removal of 50%. The NF-BWRO brine was treated in 5 L batches (a total of 21 batches), applying a constant current of 5 Ampere for 587 s (see Equation (7)). Oxygen was supplied by an air pump, while rapidly mixing the solution. This was followed by a 22 RPM and 25 min slow mixing. The solution was left to settle overnight.

Samples were taken from each batch, and the supernatant and sludge were collected. The sludge was separated using a centrifuge at 3000 RPM for 10 min. All the combined supernatants were filtered using a GF/A filter paper to simulate a UF or MF filtration step.

2.3. PHREEQC Simulation for Water Characterization

The PHREEQC software, Version 3.8.0 [32], using the SIT (sit.dat) and Pitzer (pitzer.dat) thermodynamic databases, was used to determine the saturation indices (SI) of potentially precipitating solids in both the NF and RO retentates (streams 4 and 3 in Figure 1). For computing the expected retentate SI values in the RO steps, a 100% rejection for all solutes was assumed. The Pitzer database was used to calculate the osmotic coefficients and osmotic pressure.

2.4. Retentate Evaporation Simulations

The MVR type evaporation was simulated by ChemCad NXT^® 1.1.1.17611 process simulation software (CC). The electrolyte model was applied to account for the actual solutions characteristics. The CC process simulation results were used for estimating the MVR CAPEX and OPEX.

3. Results and Discussion

3.1. Generation of the NF/RO Retentate

3.1.1. Results from the NF Step

Two NF membrane (DL type, see Table 3) filtration repetitions were carried out with 607 L of raw BW. The alkalinity of the water (5.8 meq/L) was reduced by 80% to prevent carbonate-based scaling on the membrane, by dosing HCl 32%. This was carried out to ensure that the limiting factors would become gypsum and silica precipitation; the prevention of CaCO₃ fouling is common practice in desalination plants. The filtration step was operated continuously. Samples were taken at the 90–94% RR points (1% jumps) from the feed tank (termed retentate accumulated (RA)), the permeate tank (termed permeate accumulated (PA)), and from the momentary permeate (P). It is noted that the NF step could likely be operated at 96% or higher, but in the pilot system, for technical reasons, it was decided not to extend the RR beyond 94%.

Indeed, the results, listed in

Table 4. Empirical composition of the accumulated NF (DL membrane) retentate at various recovery ratios.

RR	SO42−	SiO2	Ca2+	Mg2+	K+	Na+	SI Gypsum
%	mg/L	mg/L	mg/L	mg/L	mg/L	mg/L	-
0	428	20	201	198	44	1378	−1.52
90	3045 ± 87	27 ± 0	809 ± 4	1230 ± 4	50 ± 1	1458 ± 26	0.02
91	3274 ± 9	26 ± 0	848 ± 11	1343 ± 28	49 ± 1	1478 ± 30	0.07
92	3628 ± 136	26 ± 0	926 ± 1	1428 ± 46	50 ± 1	1443 ± 14	0.13
93	4073 ± 117	27 ± 0	995 ± 25	1574 ± 5	49 ± 0	1447 ± 48	0.20
94	4634 ± 138	28 ± 0	1134 ± 17	1791 ± 15	50 ± 0	1492 ± 18	0.26

, show that throughout the NF step, the calculated SI values were much lower than the limiting SI of gypsum, which is between 0.4 and 0.6 in the presence of antiscalants [5,33]. Further calculations revealed that a RR of 96% is also highly feasible and possibly also 97%.

The distribution of species in the accumulated permeate (

Table 5. Empirical composition of the accumulated NF (DL membrane) permeate at various recovery ratios.

RR	SO42−	SiO2	Ca2+	Mg2+	K+	Na+
%	mg/L	mg/L	mg/L	mg/L	mg/L	mg/L
90	31 ± 4	20 ± 0	122 ± 6	67 ± 2	43 ± 1	1285 ± 17
91	31 ± 1	20 ± 0	124 ± 1	70 ± 0	42 ± 0	1281 ± 40
92	31 ± 1	20 ± 0	128 ± 1	70 ± 0	43 ± 1	1252 ± 1
93	33 ± 1	20 ± 0	131 ± 0	72 ± 1	43 ± 0	1289 ± 10
94	33 ± 2	20 ± 0	128 ± 0	72 ± 2	42 ± 0	1306 ± 10

) shows only small variations with the RR, indicating that also from this viewpoint, the RR could be higher than 94%.

Considering the results from the accumulated permeate, a PHREEQC simulation was performed to compute the SI of both gypsum and CaF₂ at RR = 90% that would develop in the BWRO step, assuming 100% rejection for all the ions. The results, shown in Figure 2, reveal that 90% RR can be applied without inducing both silica- and calcium-based solids precipitation.

3.1.2. Results Derived from Operating the 1st RO Step on the NF Permeate (SW30HRLE)

The accumulated NF permeate generated in the two NF runs was placed in the feed vessel, and Vitec^TM 4000 (Avisrunsta [34]), a silica inhibitor antiscalant, was added to yield an initial concentration of ~3 mg/L. This solution was then subjected to an RO filtration step. The results of the accumulated retentate from these runs are shown in

Table 6. Empirical results of applying the (1st) RO step on the BWNF permeate. Shown are the concentrations of the major cations, sulphate, the required external pressure, and three calculations of the resulting RR, based on 100% rejection of (1) SiO₂; (2) Ca²⁺; and (3) based on retentate volume measurements.

RO RR Calculation Based On:			Na+	K+	Ca2+	Mg2+	SO42−	SiO2	B	Pressure
[SiO2]	[Ca2+]	Retentate volume	mg/L	mg/L	mg/L	mg/L	mg/L	mg/L	mg/L	bar
0%	0%	0%	1293 ± 6	44 ± 2	130 ± 0	74 ± 0.5	29 ± 0.3	19 ± 0	0.1 ± 0	20
70%	71%	70%	4248 ± 75	144 ± 2	443 ± 3	251 ± 3	109 ± 18	65 ± 1	0.1 ± 0.1	25
75%	74%	75%	5039 ± 16	172 ± 1	507 ± 5	291 ± 2	119 ± 6	77 ± 1	0 ± 0	28
80%	80%	80%	6275 ± 175	215 ± 1	635 ± 1	356 ± 4	150 ± 6	95 ± 0	0.2 ± 0.2	32
84%	84%	85%	7754 ± 35	267 ± 1	790 ± 6	446 ± 12	185 ± 6	119 ± 2	0 ± 0	37

, along with the required external pressure. A back calculation was performed to compute the RR using the Ca²⁺ and SiO₂ concentrations measured in the retentate, assuming 100% rejection for these species (a valid assumption). Using this technique, the actual calculated RR was found to be 88% (rather than 90%, calculated based on the retentate volume), the deviation likely due to the unknown solution volume left in the RO system and a small variation in measuring the initial feed tank volume. In the accumulated permeate, no SiO₂, Mg²⁺, Ca²⁺, or SO₄²⁻ concentrations were observed. The final Na⁺ concentration was ~50 mg/L, with only traces of boron (below 0.1 mg/L). The final silica concentration was 153.4 mg/L, showing that the RR could have been higher also in this step, amounting to at least 90% before exceeding the maximal allowed silica concentration in the presence of a silica antiscalant (200 mg SiO₂/L, as mentioned before). The increase in pressure was almost linear with the RR, as expected. The final TDS in the retentate solution was ~32 g/L, i.e., slightly lower than seawater, hinting that an additional 50%-RR 2nd step would be possible without exceeding the maximal external pressure (70 bar) on the RO membrane. No precipitation was visible in the accumulated retentate a week after the experiment, corroborating the no chemical fouling expectations.

3.2. Silica Co-Precipitation via Iron Electro-Dissolution

3.2.1. Determination of the Required Iron Dose

To establish the required range of iron dosage for silica co-precipitation, two jar test experiments were performed, using a synthetic retentate solution, various iron dosages, and either ED or a concentrated FeCl₃ solution as the Fe supply source.

The results of the two experiments are presented in Table S1 and

Table 7. Results of the 2nd silica co-precipitation experiment (operational conditions are listed in the Section 2).

Iron Dose		FeCl3 Dose		Fe Electro-Dissolution
	% removal		Current	Time	% removal
mgFe/L	AVG	STDEV	A	s	AVG	STDEV
80	29%	0.5%	4	41	45%	1.2%
110	35%	2.8%	4	57	54%	4.3%
140	41%	3.0%	4	73	60%	1.5%
170	49%	4.9%	4	88	66%	4.6%

. Both methods yielded similar results, but the ED method achieved higher efficiency, making it preferable, because it did not add chlorides, and its overall reaction (Equations (2)–(6)) did not consume alkalinity, thereby maintaining the pH stable.

From the results in Table S1 and Table 7, it was concluded that a dosage of at least 110 mgFe/L is required to ensure the target 50% reduction in the SiO₂ concentration in the NF/RO retentate.

The sludge that was produced in the second experiment was weighed dry. The results, presented in Table S2, show a clear increase in weight at the higher dosages, with relatively low STDEV values, which corroborated the repeatability of the method.

3.2.2. Treating the NF/RO Retentate to Attain at Least 50% Reduction in the Silica Concentration

The produced NF/RO retentate (Table 6) was treated to remove SiO₂ using the electro-dissolution method. The analysis of 21 treated 5 L batches indicated an average silica concentration reduction of 79% ± 1.6%, resulting in a final silica concentration of ~32 mg/L. Since the maximum allowable silica concentration for the RO step is 200 mg/L, this treatment allows for an additional RO desalination step of >50% RR, in which the osmotic pressure that would develop in the retentate of this step would become the limiting operational factor (rather than the silica concentration).

3.3. Results from Operating the 2nd RO Pass on the Treated NF/RO Retentate

Sixty liters of the ED-treated retentate (with a low SiO₂ concentration) were subjected to a 2nd-stage RO, using the same membrane (SW30HRLE). A constant permeate flow rate was set by gradually increasing the external pressure, and samples were taken from the accumulated permeate and retentate at the following RR values: 17%, 25%, 33%, 42%, and 50%. The required pressure at each point was recorded. The water temperature was maintained at 25 ± 2 °C.

The results, presented in

Table 8. Results of the RO step performed on the ED treated NF-RO combined retentate. The silica and major ion concentrations are shown as a function of the RR and the osmotic and applied pressures.

RR	RR	SiO2	Ca2+	K+	Mg2+	Na+	SO42−	Cl−	Calculated Osmotic Pressure	Applied Pressure
By volume	by Ca	mg/L	mg/L	mg/L	mg/L	mg/L	mg/L	mg/L	bar	bar
0	0	28 ± 0	939 ± 8	350 ± 6	500 ± 3	8942 ± 57	226 ± 1	17,057 ± 0	25.1	40
17%	13%	34 ± 0	1078 ± 7	396 ± 5	587 ± 4	10,584 ± 105	264 ± 3	20,101 ± 188	29.9	45
25%	20%	36 ± 0	1177 ± 6	433 ± 2	637 ± 3	11,386 ± 51	288 ± 2	21,677 ± 93	32.5	48
33%	29%	40 ± 0	1312 ± 4	485 ± 1	707 ± 2	12,583 ± 49	316 ± 4	23,992 ± 87	36.2	51
42%	35%	44 ± 1	1452 ± 15	539 ± 7	778 ± 8	13,818 ± 141	348 ± 4	26,376 ± 269	40.2	55
50%	43%	50 ± 1	1642 ± 18	613 ± 8	876 ± 8	15,542 ± 199	392 ± 4	29,691 ± 367	45.9	60

, show that an additional 50% RR is possible, since the osmotic pressure (OP) of the retentate at the end of the experiment was lower than 50 bar (a typical SWRO final retentate OP). Here, again, a deviation was recorded between the RR calculated using the holding vessel volume and the value calculated from the Ca²⁺ concentration in the retentate. As mentioned before, this was probably due to errors in pinpointing the initial solution volume. Nonetheless, since the SI of gypsum was negative throughout the process, the required initial silica concentration could be attained by the ED pretreatment step, and the osmotic pressure was lower than 50 bar, the assumption that the process could be applied with an RR of 50% was thus corroborated.

3.4. MVR-Based Evaporation/Condensation Stage

The combined BW desalination NF+2xRO membrane scheme described thus far would reach a recovery of 87.4% of the BW feed flow. To increase the recovery to >95%, a mechanical vapor recompression evaporation (MVR) was simulated on the mix of the retentates from the NF and 2nd RO stages. Combined, this stream had an average TDS of ~4.5% wt., which is higher than seawater but certainly common as feed or the circulation of industrial wastewater evaporation or BW advanced recovery stages [35,36]. Figure 3 shows a schematic general design for the MVR evaporation stage, planned for recovering 204.6 t/h of product water, to elevate the recovery of a 20 Mm³/y (2670 m³/h at 90% operability) BW desalination plant to >95% (i.e., 2537 m³/h product water). The main MVR equipment characteristics are shown in the Supplementary Material.

The feed (combination of the NF and 2nd RO retentates) is pH-adjusted to pH 8.5 and is preheated first by the product condensate and then by the concentrated brine residue. It is then fed to the brine circulation over the evaporator flash tank (V-1004) and the evaporator (plate heat exchangers H-1005) cold side. The water vapor from the evaporator flash tank is compressed to 0.4 barg by a centrifugal steam blower, de-superheated by partial recycling of the product water, and fed to the hot side of the evaporator. The centrifugal steam blower would consume 19 kWh/m³ product, which is well within the common practice of low-pressure steam blowers’ energy consumption. A 3% supplementary steam is required to compensate for the system’s minor inefficiencies. The product water, which is characterized by very low electrical conductivity (EC), is combined with the 1st RO product water that would then undergo post-treatment, prior to supply.

3.5. Cost Analysis

The suggested process scheme, with its relevant flow rates, RR values, and main equipment characteristic sizes, is presented in Figure 4. RO1 and RO2 in Figure 4 are the 1st and the 2nd reverse osmosis applied steps.

A full cost analysis (CAPEX and OPEX) was carried out on the proposed treatment scheme for a 20 Mm³/y design-basis scenario, to assess the cost of each separate step and the real-life feasibility of the ZLD approach. The characteristic size of each unit operation was calculated according to the treatment mass balance, experimental parameters, and common-practice general sizing procedures.

The total surface of the membranes and the pumps’ discharge pressure were calculated according to the experimental fluxes and working pressures, respectively. Pressure exchangers were sized according to an assumed membrane pressure drop of 4 bar and an efficiency of 95%. The electro coagulator (R-1000) was sized according to the experimental current density of 80 mA/cm² and 10 min hydraulic residence time, the faradaic efficiency of 100% [37], and a Re of 50–100. An 80 min residence time [38] was assumed in the settling tank (ST-1200). The pressure filter (F-1300) was sized according to an assumed metal hydroxide slurry characteristic of 2% solids at the settling tank bottom [39] and a commercially available pressure filter [40], for a feasible 4 h duration between filter cake discharges. The MVR equipment sizing was a part of the CC simulation. The MVR full required capacity was divided into three parallel systems, thereby increasing feasibility and operability. Three holding tanks with an estimated residence time of 15 min were planned for the electrocoagulation, MVR, and the product solution. The proposed system’s CAPEX estimation details are shown in the Supplementary Material. The CAPEX calculation, including its various components (materials, civil engineering, design, etc.), was based on the key equipment cost, which was factored using typical percentage of fixed-capital investment for cost segments [41]. The equipment cost estimation was based on real budget quotations and on an estimation of purchased costs of process equipment, according to specific cost estimation equations, detailed in Couper (2009) [42]. Where needed, purchase costs were time-updated according to the chemical engineering plant cost index (CEPCI) from October 2024. The referenced quotation dates are listed in the Supplementary Material for each piece of equipment.

The purchased equipment costs, calculated in the Supplementary File, were used to estimate the CAPEX of the proposed desalination plant.

To enable an incremental cost estimation, the CAPEX was calculated separately for the membranes and the MVR sections of the proposed plant. Thus, the equipment estimation of the membranes section was factored as 23% of the total membrane CAPEX [41], while the key equipment of the MVR section, which included costly high-alloy heat exchangers and expensive machinery, was factored as 40% of the total MVR CAPEX. The specific CAPEX in USD /m³ was calculated with an interest of 4.375% [43] for 30 years [44]. However, utility projects in the US may receive discounted rates of about 2.5% or even as low as 1.6% interest rate for long term projects in eligible municipalities [45], which may reduce the total product cost (CAPEX + OPEX) by 4–5%.

The CAPEX at our reference brackish water desalination plant in Ma’agan Michael cost USD 700/(m³ product/d) [S. Oren, personal communication, 16/1/2025], from which 6% was deducted to account for the post-treatment CAPEX, which was not included in the current cost comparison.

The estimated CAPEX of the various options and the estimated capitalized investment (4.375%, 30 y) per m³ of product water and USD /(m³/d), for 20 Mm³/y, are summarized in

Table 9. Proposed plant CAPEX calculations.

	Purchased Equipment Cost	Equipment for CAPEX Factor	Estimated CAPEX	Normalized Investment [USD /m3]	Normalized Investment [USD /m3/d)]
NF/RO1 baseline (82.7% RR)	MUSD 6.953	23%	MUSD 30.23	0.0953	540
NF/RO1/ND/RO2 (87.4% RR)	MUSD 9.017	23%	MUSD 39.21	0.1236	682
MVR (95% RR)	MUSD 8.641	40%	MUSD 21.60	0.7762	4400
Total plant			MUSD 60.81	0.1762	999
Existing Ma’agan Michael reference (82% RR)					658 (1)

⁽¹⁾ The Ma’agan Michael CAPEX after 6% deduction for the post-treatment.

.

A separate estimation of the MVR section, using a budgetary price of EUR 4,500,000 for a 15 t/h brackish water MVR and the six-tenths-factor rule [41], yielded an estimated cost of USD 22.82 million for the proposed 204.6 t/h MVR, which agrees with the presented equipment-based estimation. Similarly, the membrane section CAPEX estimation, which included the initial NF stage and the silica removal addition, agreed with the USD 700 per m³/d of the Ma’agan Michael ballpark CAPEX indication.

The results of the OPEX analysis are presented in the Supplementary Material, which lists the expected operation costs of the membrane in the baseline case (i.e., NF/RO1, RR = 82.7%), the increased RR case (NF/RO1/ED/RO2, RR = 87.4%), and the full ZLD case (NF/RO1/ED/RO2/MVR, RR = 95%), also in comparison with the Ma’agan Michael plant reference case (2 RO stages, RR = 82%).

For the OPEX assessment, the following dataset was used: The cost of the iron plates was assumed at USD 700/ton [46]; the estimated power consumption of the various pieces of equipment are detailed in

Table 10. Collective OPEX and CAPEX data and incremental normalized costs analysis.

Option	Total Flow [m3/h]	Incremental Product [m3/h]	OPEX [USD /m3]	Recovery Ratio	Total Cost [USD /m3]	Incremental Total Cost [USD /m3]
Two-stage RO (Ma’agan Michael plant’s actual operation) (1)	2100	0	0.338	82.0%	0.445	----
NF/RO1	2209	108.6	0.332	82.7%	0.439	0.439
NF/RO1/ED/RO2	2332	143.1	0.408	87.4%	0.531	1.160
NF/RO1/ED/RO2/MVR (Full ZLD solution)	2537	204.6	0.526	95.0%	0.702	2.647

⁽¹⁾ The reference plant on the same basis as the proposed plant.

; the electricity cost was assumed at USD 0.0872/kWh [47]; the Genesys CAS antiscalant was used for the NF and the 1st RO stage at a concentration of 1 mg/L in the feed. Its cost was estimated at USD 2000/ton [33], whereas the Vitec 4000 antiscalant, used to prevent silica precipitation in the 2nd RO stage at a concentration of 5 mg/L in the feed [34] was estimated at USD 12,950/ton [48]; brine disposal via deep well injection was estimated at USD 0.54/m³ [16]. Four employees were assumed at the membrane section in the morning shift and one at the MVR section. Two employees were assumed at the MVR section in the night shift; maintenance was assumed at 3% of the CAPEX for the reference plant in Ma’agan Michael [49], while 5% and 2% of the plant CAPEX were assumed for the membrane and the MVR sections of the proposed plant, respectively. Among other maintenance expenses, membrane replacement was assumed every 3 to 7 years, which is a significant continuous maintenance cost; sludge disposal to landfill was estimated at USD 0.11/kg [50]; the solids content in the ED sludge was assumed to be 20% (w/w); the backwash of the UF filtration was estimated at 5% of the feed flow rate. The OPEX analysis was based on a capacity of 2332.7 and 204.6 m³/h in the membrane and MVR sections, respectively, in the proposed plant, and 2100 m³/h in the Ma’agan Michael reference plant. A turbocharger (TC) with an efficiency of 60% was used in the Ma’agan Michael (MM) reference plant in both RO stages that operate at RR of 66% and 47%, along with a 4-bar pressure drop in each filtration stage.

The breakdown of the OPEX and the return on investment value for the proposed ZLD plant are shown in Figure 5.

Table 10 and Figure 6 summarize the OPEX and CAPEX of the operational options (recovery ratios) and the incremental total cost for the proposed plant’s three operational options (as well as for the Ma’agan Michael reference plant). The red triangle in Figure 6 stands for the normalized total cost of the Ma’agan Michael reference plant. The actual brine disposal solution and its associated costs have a critical role in the difference between the three operational options. For example, an offsite brine injection, carried out with tankers that may cost at least USD 3.5/m³ [51], will render the full ZLD option the most economic, with a total normalized cost of USD 0.846/m³ versus 1.013 and 1.095 USD /m³ in the baseline operation and the Ma’agan Michael reference case, respectively.

Considering that the overall RR was 95.0%, as compared to 82.0% in the currently applied desalination scheme and 82.7% in the NF/RO1 baseline case, the limiting parameter for increasing the recovery in the 2nd-stage RO step, following the SiO₂ precipitation, became the osmotic pressure that developed in the 2nd pass RO retentate. The application of MVR, despite being relatively expensive, as it contributes ~32% of the overall product cost, is a readily available method for increasing the RR to ≥95%. Other evaporation-based methods could, clearly, be used in this final step, but if used, based on thermodynamics, the overall cost increment would not change significantly, and the conclusions of this study, with respect to the normalized cost of each step, would probably still hold.

4. Conclusions

Attaining ZLD in inland BW desalination is desired and challenging at the same time. Many suggestions exist in the literature but hardly any of them is implemented in reality. In this work, we combined several proven technologies (NF, RO, Fe electro-dissolution, MVR) in a novel fashion, to yield RR of 95%, while assessing their related costs. The results show that increasing the RR from 82.7% to 87.4% increased the normalized total cost by 21% (USD 0.445/m³ to USD 0.531/m³), while further increasing the RR to 95.0% resulted in a further 32% increase, to USD 0.702/m³ of product water (full flowrate basis). The actual brine disposal solution and its cost has a critical role in the economics of the selected desalination option. Obviously, the ZLD option is more favorable when the brine disposal cost increases. When the silica concentration in the feed is ~20 mg/L (common in many groundwaters), the bottleneck of increasing the RR to beyond ~88–90% becomes the silica concentration that develops in the retentate to reach >200 mg/L. This SiO₂ concentration must be reduced considerably to allow applying the final desalination step. In this work, we chose to apply Fe electro-dissolution to remove the silica. Whatever the chosen method is, this step is indeed costly but cannot be circumvented. Nevertheless, the results show that attaining >95% recovery is feasible at a reasonable cost, that in many places inland may be justified, relative to other possible alternatives. The accuracy of the cost estimation, according to the principles listed in [52], is >80% (i.e., <±20%). The CAPEX error is estimated at ±30% (Class 4) and OPEX at ~10%.