Reverse osmosis (RO) now accounts for roughly 66-69% of installed global desalination capacity, and in most trade coverage that number gets read as a verdict: membranes won, thermal lost. The picture is less settled than that. Judged by second-law (exergy) efficiency, multi-effect distillation (MED) is the more thermodynamically complete process of the three mainstream technologies — and in the Gulf Cooperation Council states specifically, thermal desalination (MED plus multi-stage flash, MSF) still holds close to 85% of the market. That split isn’t inertia. It reflects a real difference in what each process is actually being asked to do: RO buys cheap electricity and fights osmotic pressure with a pump; thermal desalination buys cheap (often otherwise-wasted) steam and fights nothing but its own heat-transfer losses. Which one wins is a question about the energy market and feedwater chemistry at a specific site, not a question with a single global answer.
This piece works through the physics each process is built on, why their pretreatment requirements diverge by roughly an order of magnitude, how to compare their energy consumption on a common basis despite one running on electricity and the other on steam, and what three real full-scale plants — Sorek II in Israel, Taweelah in the UAE, and the hybrid Ras Al Khair complex in Saudi Arabia — actually did with these tradeoffs.
The energy floor nothing can beat
Desalination is, thermodynamically, unmixing — separating a dilute salt solution back into pure water and a more concentrated brine. That’s an entropy-decreasing process for the water/salt system, so it requires a minimum external energy input regardless of which technology performs the separation. For an ideal dilute binary solution of water and salt, the Gibbs free energy of mixing is:
where is total moles, the gas constant, absolute temperature, and , the mole fractions of water and salt. Because seawater is dilute in the thermodynamic sense ( is small), this simplifies via a Taylor expansion to:
Normalizing the reversible separation work to the volume of product water gives the theoretical minimum specific energy consumption (). At standard seawater salinity and 25°C, depends on the recovery ratio the system is targeting — the fraction of feed converted to permeate. In the limit of recovery approaching zero (an infinitesimally small draw from an effectively infinite reservoir), . At a recovery ratio of 50% — which is closer to what real plants actually run — the residual brine has concentrated enough that its osmotic pressure has risen substantially, and climbs to roughly .
That’s the floor. No amount of engineering gets a real plant to this number, because real systems also have to pay for pump and turbine hydraulic losses, membrane channel friction, boundary-layer mass-transfer resistance, and non-equilibrium heat transfer — none of which are optional. What the floor is useful for is context: when you see an RO plant quoting 2.5-4.0 kWh/m³, that’s roughly 2-4x the reversible minimum at realistic recovery, and closing that remaining gap is almost entirely what a century of desalination engineering has been about.
How RO actually pulls water through a membrane
RO is pressure-driven membrane separation: applied hydraulic pressure exceeding the feedwater’s osmotic pressure forces water molecules against their chemical-potential gradient, through a dense, effectively nonporous polyamide active layer, while dissolved ions are rejected on the feed side.
The dominant mechanistic model is solution-diffusion: solvent and solute are treated as dissolving into the membrane matrix and diffusing across it independently, driven by different gradients. Water flux follows the net effective pressure differential:
where is the membrane’s water permeability coefficient, the applied pressure differential, and the osmotic pressure differential across the membrane. Critically, solute (salt) flux is governed by a completely different driving force — the concentration gradient alone, independent of applied pressure:
where is the membrane’s solute permeability coefficient and , are the solute concentrations at the membrane surface and in the permeate, respectively. This decoupling is the reason RO salt rejection doesn’t improve by simply running the plant at higher pressure — pushing more water through doesn’t slow the salt down, it just dilutes it into more permeate.
The complication is concentration polarization: because salt is continuously rejected at the membrane surface, a boundary layer forms there with salinity substantially higher than the bulk feed. This inflates the effective term right where it matters most, cutting into the driving force for , and it also raises the local risk of sparingly-soluble salts crystallizing out at the membrane surface before the bulk solution is anywhere near saturation.
How MED and MSF pull water out by boiling, not squeezing
MED and MSF are both thermal processes, but the way they extract latent heat is different enough that they behave like different technologies in practice, not variations on a theme.
MED cascades vaporization and condensation across a series of evaporator vessels (“effects”) run at successively lower pressure. Motive steam condenses in the first effect’s tube bundle, boiling sprayed seawater on the shell side; the vapor that boils off is slightly cooler than the motive steam, so the second effect is held at lower pressure so that vapor can still boil the feed there too — and so on down the train, with the final effect’s vapor condensed against incoming cold feed. Each unit of input steam energy effectively does its condensation/evaporation work multiple times as it cascades down the pressure ladder, which is the entire efficiency argument for MED.
MED’s top brine temperature (TBT) is deliberately capped, typically in the 65-70°C range. That’s not an equipment limitation so much as a scale-chemistry one — pushed hotter, alkaline scaling accelerates sharply.
MSF works differently: it doesn’t rely on boiling against a heat-transfer surface at all. Preheated, pressurized seawater is heated in a brine heater to its TBT (typically 90-110°C), then flashed — released stage by stage into chambers held at successively lower pressure. At each stage the liquid is momentarily superheated relative to the local saturation pressure, and that excess sensible heat converts almost instantaneously into flash vapor. Because residence time in each flash chamber is short, the fluid rarely reaches true thermodynamic equilibrium before moving to the next stage; the resulting temperature gap is the Non-Equilibrium Allowance (NEA), and it’s the single largest source of thermodynamic loss in MSF specifically — a loss mechanism RO and MED simply don’t have an equivalent of.
A newer, still-niche entrant worth flagging: thermodiffusive desalination (TDD) exploits the Soret effect — thermal diffusion of dissolved ions along a temperature gradient — to desalinate without any phase change at all. A single TDD pass is a modest desalting step, but cascaded stages can reach deep desalination, and because there’s no boiling and no high-pressure pump, electrical consumption is extremely low. That combination (low electrical demand, tolerance of low-grade heat, no high-pressure hydraulics to maintain) makes TDD a plausible fit for remote, off-grid islands with abundant low-grade heat and essentially no reliable grid power — not a competitor to RO/MED/MSF at utility scale.
Why pretreatment requirements differ by an order of magnitude
RO’s fouling sensitivity comes directly from its physics: anything that deposits on or in a dense polymer membrane degrades flux and rejection permanently, and there’s no thermal self-sterilizing effect to fall back on. Typical RO feed specifications require a Silt Density Index and sustained turbidity below 0.1 NTU. Meeting that reliably has pushed most large modern SWRO plants away from conventional media (sand) filtration and toward ultrafiltration (UF) as the pretreatment backbone — UF pore sizes in the 0.01-0.03 μm range provide a hard physical barrier against colloids, viruses, bacteria, and microalgae that media filtration can’t guarantee.
Nanofiltration (NF) is increasingly used ahead of both RO and thermal systems specifically for hardness removal: NF’s charge-based (Donnan) rejection selectively strips divalent scale-forming ions (calcium, magnesium, sulfate) from the feed, which lowers the calcium sulfate and similar scaling risk downstream and allows the subsequent RO or thermal stage to run safely at a higher limiting recovery ratio. Readers modeling this pretreatment step — independent monovalent/divalent rejection tracking across a multi-stage NF array — can work the mass balance directly with this site’s NF System Designer.
Chemically, RO feed needs continuous antiscalant dosing (typically phosphonate or polyacrylate-based, to raise the nucleation barrier for sparingly-soluble crystals) and dechlorination: free chlorine and other strong oxidants attack the polyamide active layer directly, degrading the polymer and collapsing salt rejection, so sodium metabisulfite (SBS) is dosed ahead of the membrane train to hold oxidation-reduction potential below a safe threshold. Readers sizing these dosing rates can use this site’s Chemical Dosing Calculator.
Thermal processes tolerate feedwater turbidity that would foul an RO train in hours — coarse screening plus self-cleaning strainers is often sufficient. Their real bottleneck is high-temperature scale and corrosion chemistry. Heating seawater to thermal-process temperatures drives thermal decomposition of bicarbonate:
The carbonate produced combines with calcium to precipitate hard, poorly heat-conductive calcium carbonate scale directly on heat-transfer surfaces. At higher operating temperatures, water’s dissociation equilibrium shifts, hydroxide concentration rises, and magnesium hydrolyzes to precipitate flaky magnesium hydroxide scale:
To keep heat-transfer performance from degrading, thermal plants run continuous high-temperature-tolerant antiscalant dosing and vacuum deaeration to strip dissolved oxygen and acid gas (dissolved oxygen is typically driven down to trace levels), protecting carbon-steel and copper-nickel alloy tube bundles from pitting and general oxidative corrosion at temperature.
Comparing energy consumption when one process runs on electricity and the other on steam
RO consumes almost exclusively high-grade electrical work; MED and MSF consume mostly low-grade thermal energy (plus a modest electrical load for pumps and vacuum systems). Comparing them honestly requires converting both to a common basis — the Standard Primary Energy (SPE) framework does this by tracing each energy input back to the primary fuel required to produce it. Electrical energy is converted using a conversion factor , based on a roughly 50% generation efficiency benchmark (a modern combined-cycle gas turbine plant). Low-pressure industrial steam below 130°C is converted using — a number that looks large because it is: low-grade heat can’t be converted back to electrical work with anywhere near the same efficiency a generator achieves going the other direction.
Within RO’s own electrical consumption, the biggest single efficiency lever has been energy recovery. High-pressure pumps push feed to roughly 55-75 bar; at a typical 40-45% recovery, the 55-60% of feed that leaves as high-pressure concentrate still carries most of that hydraulic pressure energy. Without an energy recovery device (ERD), that pressure is simply throttled away as friction heat, and specific energy consumption runs 8-12 kWh/m³. Centrifugal-type ERDs (turbocharger-style devices) recover concentrate energy via a rotor, converting hydraulic pressure to rotational mechanical work and back to hydraulic pressure on the feed side — two conversions, each with friction, leakage, and impact losses, capping practical transfer efficiency around 80-85%. Isobaric (rotary pressure exchanger) devices instead bring high-pressure concentrate and low-pressure feed into direct contact within a rotating ceramic-cartridge rotor, transferring pressure with no intervening shaft-work stage; transfer efficiency runs 95-98%. That efficiency gain is the single biggest reason modern SWRO specific power consumption (SPC) has fallen from roughly 10 kWh/m³ a generation ago to a typical 2.5-4.0 kWh/m³ today.
Judged purely on kWh consumed, RO wins by a wide margin — it has essentially no thermal energy line item at all. But judged on second-law (exergy) efficiency and SPE, the ranking is closer, and in one reading, inverted: MED’s efficient use of low-grade cascade heat gives it a second-law efficiency around 14.2%, against roughly 11.1% for RO. The practical implication is specific rather than universal: if a site has access to genuinely low- or zero-cost heat (industrial waste heat, a nuclear plant’s condenser reject heat), MED’s primary-energy economics can outperform even a state-of-the-art isobaric-ERD RO train. MSF, by contrast, loses on every efficiency metric in this comparison — its inherent non-equilibrium flash losses produce the largest entropy generation of the three processes, and it is not the process to specify for a new low-carbon build unless cheap thermal energy is already unavoidable on site.
What this costs to build and run
RO’s capital structure is light and modular: standardized membrane elements and fiberglass pressure vessels bolt together quickly, civil works are largely limited to intake pumping and UF basins, and capital spend concentrates in high-value mechanical equipment (high-pressure pumps, isobaric ERDs, duplex stainless piping). MSF sits at the other extreme — its flash-chamber vessels and condenser tube bundles consume large volumes of expensive corrosion-resistant alloy (Hastelloy, copper-nickel, or titanium), site assembly is slow, and construction cost commonly runs more than double an equivalent RO plant. MED, by running cooler, can use cheaper anti-corrosion coatings and thin, efficient titanium plate, landing its CAPEX between the two — well above RO, well below MSF.
Operating cost tells a related but distinct story. RO membrane elements are consumables — a full-plant replacement cycle every 3-5 years — and membrane/dosing chemical costs typically run 15-20% of OPEX; RO OPEX is also unusually exposed to electricity price volatility, since power commonly makes up 40-50% of running cost. Thermal plants have far fewer wear parts and correspondingly lower consumables spend (well under 5% of OPEX), but they’re just as exposed — arguably more so — to the price of the steam they consume.
| RO | MED | MSF | |
|---|---|---|---|
| Typical design life | 15-20 yr (cyclic high pressure, membrane aging) | 25-30 yr (lower thermal stress) | 35+ yr (heavy, mechanically simple) |
| Major consumable | Membrane elements, 3-5 yr replacement | Minimal | Minimal |
| Consumables share of OPEX | 15-20% | <5% | <3% |
| LCOW | $0.45-1.72/m³ | $0.55-0.95/m³ (with low-cost heat) | $1.10-2.20/m³ |
RO’s low specific power consumption is what makes it the commercially dominant choice almost everywhere power is priced normally. MED only closes the LCOW gap — and can undercut RO — where heat is close to free.
Three plants, three different bets
Sorek II (Sorek B), Israel — 670,000 m³/d, with a water price around $0.40-0.45/m³ that set a new low benchmark for large-scale SWRO at the time it was built. The engineering choice that mattered most: the plant moved from the conventional 8-inch horizontal membrane element arrangement to 16-inch elements in a vertical orientation — a single 16-inch element carries roughly four times the effective membrane area and throughput of an 8-inch unit, which cut the total count of membrane housings, flanged connections, and high-pressure piping runs, shrinking the RO building footprint and CAPEX substantially. The intake and outfall used long-distance pipe jacking rather than open-trench seabed excavation, avoiding disturbance to the seabed and coastal habitat along the route. Product boron was held below the tight threshold required for unrestricted agricultural irrigation reuse — a real test of a two-pass membrane configuration’s rejection performance for a specific problem ion, not just bulk TDS.
Taweelah, UAE — 909,200 m³/d, one of the largest single SWRO facilities built to date, and notable less for its size than for what it replaced: the site previously ran MSF units selling water at roughly $0.84/m³ with a heavy fossil-fuel combustion footprint. The rebuild standardized on isobaric energy recovery throughout and, working against summer feedwater conditions around 35°C and TDS above 45,000 mg/L — a genuinely difficult combination — still held specific power consumption to about 2.7 kWh/m³, a roughly 75% energy-intensity reduction versus the MSF baseline it replaced. That’s a fairly direct empirical answer to “how much does switching from thermal to membrane actually save at Gulf feedwater conditions.”
Ras Al Khair, Saudi Arabia — a hybrid complex combining eight MSF units (roughly 300,000 m³/d combined) with seventeen modern SWRO trains (roughly 730,000 m³/d combined), for a total capacity near 1.03 million m³/d. What makes it a genuinely different design pattern rather than just two plants sharing a fence line is the product blending: MSF condensate is nearly ion-free distillate, while single-pass RO permeate in summer can approach the higher end of acceptable finished-water salinity as membranes swell with feed temperature. Blending the two product streams in the right ratio hits target finished-water hardness and boron without adding a second RO pass at all. The complex also uses turbine condenser reject heat to gently preheat winter RO feedwater — cold intake water otherwise depresses water flux and forces higher operating pressure — smoothing the RO train’s power draw across the year.
Environmental discharge: a different failure mode per process, not a shared one
Every desalination technology returns a brine stream to the receiving water, and the three technologies produce meaningfully different discharge hazards.
RO concentrate stays close to ambient temperature but reaches roughly 1.5-2x feed salinity. Because it’s denser than the receiving seawater, it sinks and spreads as a hypersaline plume that hugs the seabed rather than mixing into the water column, creating localized hypoxia that’s particularly damaging to benthic invertebrates, seagrass beds, and demersal fish eggs — organisms with limited ability to relocate away from the plume.
MED and MSF concentrate is less hypersaline (roughly 1.2-1.5x feed salinity) but is discharged noticeably warmer than ambient — a thermal pollution problem rather than a salinity one. Elevated discharge temperature depresses local dissolved oxygen and can favor harmful algal bloom conditions, shifting the receiving ecosystem’s seasonal biological rhythm rather than its baseline chemistry.
Both process families also carry chemical residuals worth accounting for separately from the bulk brine: RO’s periodic membrane cleaning-in-place cycles discharge organic-load citric acid and EDTA-based chelant waste; MSF’s high-temperature, mildly acidic operating environment causes measurable metal loss from copper-nickel/brass tube bundles, adding chronic-toxicity copper and nickel to the brine stream.
On carbon footprint specifically, the two energy inputs diverge sharply once you price in where the energy actually comes from. Using representative grid-average and fossil-boiler emission factors as a worked example:
The message isn’t “thermal is dirty” in absolute terms — it’s that MED’s carbon footprint is almost entirely a function of where its steam comes from. Fed by a dedicated fossil boiler, its footprint is dramatically worse than grid-powered RO; fed by genuinely free waste heat, the calculation above collapses toward near-zero. That’s the same conclusion the SPE energy-accounting section reaches from a different direction: MED’s real-world performance, both economic and environmental, is inseparable from whether the site already has heat to spare.
Coupling with renewables changes the constraints, not just the fuel
Photovoltaic-driven RO (PV-RO) is the most commercially mature renewable-desalination pairing, but it runs into a real control problem: SWRO trains were designed for continuous, constant-pressure operation, and PV/wind output is inherently intermittent. If feed pressure drops below the osmotic pressure of the current feed salinity during a cloud passage or lull, permeate flow stops instantly; concentration-polarization-layer salt has no flushing flow to disperse it and can crystallize directly on the membrane surface, with severe pressure transients risking physical membrane damage.
Battery buffering was the obvious first answer, and large-format lithium battery banks can hold pressure constant through short-term output dips — but full lifecycle-cost analysis at utility scale generally kills the approach: battery CAPEX erodes RO’s cost advantage, and a battery’s 5-8 year degradation cycle is badly mismatched to a desalination plant’s multi-decade asset life, with disposal of retired battery packs adding its own environmental liability. Current engineering practice has moved toward battery-free flexible operation instead — dynamically staging parallel membrane trains in and out of service as available power changes, keeping the trains that stay online at their optimum flow velocity and rejection pressure rather than running the whole array at a diminished, off-design condition — combined with storing product water rather than storing electricity (elevated freshwater tanks filled at full tilt during peak generation, gravity-fed during lulls), which sidesteps the battery problem at the storage-medium level rather than solving it at the battery-chemistry level.
Solar-thermal coupling with MED is a more natural physical match: parabolic trough or linear Fresnel collectors heating a working fluid above roughly 150°C can drive a small organic Rankine cycle turbine for power generation, with the turbine’s rejected low-grade heat (typically 70-80°C) feeding directly into an MED first effect — a genuine heat cascade rather than two systems sharing infrastructure. Molten-salt or pressurized hot-water thermal storage in the collector loop lets the MED train run around the clock rather than only during sunlight hours, which also avoids the scaling and corrosion penalty that comes from repeated thermal cycling on start-stop operation.
The logical endpoint of stacking these tradeoffs is a cascaded hybrid: PV-RO handles the first desalting pass at standard recovery, NF or a second high-pressure RO pass strips hardness and concentrates the reject further, solar-thermal-driven MED or membrane distillation handles a further thermal concentration step once hardness-forming ions are largely gone (so the higher operating temperature doesn’t immediately trigger scaling), and a final evaporator/crystallizer recovers the remaining water as distillate while precipitating the dissolved solids as saleable sodium chloride, sodium sulfate, magnesium sulfate, and calcium carbonate. That sequencing — hardness removal before the high-temperature stage, not after — is what makes a zero-liquid-discharge cascade survivable rather than a scaling problem waiting to happen. Readers modeling the RO/NF stages of that cascade can work the recovery, concentrate TDS, and osmotic-pressure checks with this site’s RO System Designer alongside the NF System Designer linked above. For a deeper look at how that membrane-to-thermal handoff decision plays out in industrial ZLD specifically — including MVR, membrane distillation, and electrodialysis as the finishing steps — see this site’s zero liquid discharge sequencing piece.
What this means for technology selection
None of the three processes is a categorically “better” choice — each wins under a specific, identifiable set of site conditions. RO is the right default almost everywhere: its low specific power consumption, small carbon footprint per cubic meter on a normal grid, and modular light-CAPEX construction make it the lowest-risk choice for municipal supply and distributed coastal projects where electricity is priced normally and there’s no free heat sitting around. MED is worth a serious second look specifically where a site already has low- or zero-cost thermal energy — industrial waste heat, a co-located power plant’s condenser reject heat, concentrated solar-thermal input — because that’s the one condition under which its superior second-law efficiency translates into an LCOW that can beat RO outright. MSF is increasingly hard to justify for new builds: its combination of high entropy-generating losses, heavy consumption of expensive corrosion-resistant alloy, and near-total dependence on cheap fossil steam puts it at a structural disadvantage against both alternatives, and it shows up mostly in refurbishment or hybrid retrofit contexts rather than new specifications.
The practical takeaway for anyone actually specifying a plant: don’t start from “which technology is best” — start from what energy is actually available at the site, at what real price, and what the feedwater salinity and temperature profile looks like across the seasons. The technology choice mostly falls out of those three answers.
Further reading
- Fritzmann, C. et al., “State-of-the-art of reverse osmosis desalination,” Desalination 216 (2007) — the standard reference review for RO solution-diffusion modeling and membrane transport.
- El-Dessouky, H.T. & Ettouney, H.M., Fundamentals of Salt Water Desalination, Elsevier (2002) — textbook treatment of MED and MSF thermodynamics, including non-equilibrium allowance and gained-output-ratio derivations.
- Mistry, K.H. et al., “Entropy generation minimization of desalination technologies,” and the related derivation “Derivation of the Theoretical Minimum Energy of Separation of Desalination Processes,” Journal of Chemical Education 97 (2020) — the theoretical-minimum-energy derivation used in this piece.
- International Desalination Association (IDA) / DesalData market reports — the primary published source for global and regional installed-capacity share figures by technology.
- U.S. DOE / NREL technical reports on energy recovery devices and desalination energy consumption benchmarks — background on isobaric vs. centrifugal ERD performance figures.