See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/228496424

Desalination by ammonia-carbon dioxide

forward osmosis: Influence of draw and feed
solution concentrations on...

Article in Journal of Membrane Science · July 2006

DOI: 10.1016/j.memsci.2005.10.048

CITATIONS READS

433 1,364

3 authors, including:

Jeffrey Mccutcheon Menachem Elimelech

University of Connecticut Yale University
84 PUBLICATIONS 3,825 CITATIONS 502 PUBLICATIONS 45,257 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Electrocatalytic membrane fabrication and modification View project

All content following this page was uploaded by Menachem Elimelech on 17 September 2017.

The user has requested enhancement of the downloaded file.

 Journal of Membrane Science 278 (2006) 114–123

Desalination by ammonia–carbon dioxide forward osmosis: Influence of

draw and feed solution concentrations on process performance
Jeffrey R. McCutcheon, Robert L. McGinnis, Menachem Elimelech ∗
Department of Chemical Engineering, Environmental Engineering Program, Yale University, P.O. Box 208286, New Haven, CT 06520-8286, USA
Received 26 May 2005; received in revised form 17 October 2005; accepted 30 October 2005
Available online 5 December 2005

Abstract
Forward (direct) osmosis (FO) using semi-permeable polymeric membranes may be a viable alternative to reverse osmosis as a lower cost and
more environmentally friendly desalination technology. The driving force in the described FO process is provided by a draw solution comprising
highly soluble gases—ammonia and carbon dioxide. Using a commercially available FO membrane, experiments conducted in a crossflow, flat-
sheet membrane filtration cell yielded water fluxes ranging from 1 to 10 ␮m/s (2.1 to 21.2 gal ft−2 d−1 or 3.6 to 36.0 l m−2 h−1 ) for a wide range of
draw and feed solution concentrations. It was found, however, that the experimental water fluxes were far lower than those anticipated based on
available bulk osmotic pressure difference and membrane pure water permeability data. Internal concentration polarization was determined to be
the major cause for the lower than expected water flux by analysis of the available water flux data and SEM images of the membrane displaying
a porous support layer. Draw solution concentration was found to play a key role in this phenomenon. Sodium chloride rejection was determined
to be 95–99% for most tests, with higher rejections occurring under higher water flux conditions. Desalination of very high sodium chloride feed
solutions (simulating 75% recovery of seawater) was also deemed possible, leading to the possibility of brine discharge minimization.
© 2005 Elsevier B.V. All rights reserved.

Keywords: Forward osmosis; Osmosis; Desalination; Concentration polarization; Internal concentration polarization; Draw solution

1. Introduction
A worldwide shortage of unimpaired freshwater has forced
the tapping of saline water sources in water-starved countries.
Current desalination technologies, such as reverse osmosis (RO),
are, however, expensive and energy intensive. Limited recovery,
typically 35–50% for seawater [1], is another drawback of RO.
The remaining liquid, now concentrated brine, is discharged into
the ocean. This is a critical environmental drawback to RO and
limits its use to coastal areas since brine from brackish ground-
water desalination cannot be disposed of inland in an economical
manner.
Forward (or direct) osmosis (FO) is a technology that may
be able to desalinate saline water sources at a reduced cost and
at high recovery. In forward osmosis, like RO, water transports
across a semi-permeable membrane that is impermeable to salt.
However, instead of using hydraulic pressure to create the driv-
∗ Corresponding author. Tel.: +1 203 432 2789; fax: +1 203 432 2881.
E-mail address: menachem.elimelech@yale.edu (M. Elimelech).
ing force for water transport through the membrane, the FO
process utilizes an osmotic pressure gradient. A “draw” solution
having a significantly higher osmotic pressure than the saline
feed water flows against the permeate side of the membrane,
and water naturally transports across the membrane by osmo-
sis. The osmotic driving forces in FO can be significantly greater
than hydraulic driving forces in RO, potentially leading to higher
water flux rates and recoveries. The lack of hydraulic pressure
may make the process less expensive than RO, while the min-
imization of brine discharge reduces the environmental impact
of the desalination process.
Previous work on osmosis through semi-permeable mem-
branes, while limited, still exposed the two significant drawbacks
of FO. Membrane technology has advanced around the prospect
of RO, not FO. Research conducted on osmotic processes con-
cluded that RO membranes were not ideal for osmosis as water
flux was found to be low [2–5]. Finding easily separable draw
solutes is the second drawback of FO. Several patents describe
different configurations for the use and removal of FO draw
solutes [6–14]. Choosing a tailored FO membrane as well as
an appropriate draw solute is critical in making the desalination

0376-7388/$ – see front matter © 2005 Elsevier B.V. All rights reserved.
doi:10.1016/j.memsci.2005.10.048
 J.R. McCutcheon et al. / Journal of Membrane Science 278 (2006) 114–123 115

process feasible. In the work presented in this paper, a mem-

brane specifically designed for FO was used and our draw solutes
and solute composition were chosen based on several important
chemical attributes.
The ideal draw solute has several characteristics. The
solute(s) must have a high osmotic efficiency, namely high sol-
ubility in water and relatively low molecular weight, which can
lead to high osmotic pressures. The solute must also be eas-
ily and inexpensively separated to yield potable water, without
being consumed in the process. We have recently demonstrated
that these criteria are satisfied by using a draw solution of highly
soluble gases—ammonia (NH3 ) and carbon dioxide (CO2 ) [15].
These gases are highly soluble in water and have a low molecular
weight while also being relatively easy to remove from water. We
have shown that by proper management of ammonia and carbon
dioxide ratios and ammonium salt speciation, we can generate
Fig. 1. SEM image of a cross-section of the forward osmosis (CA) membrane. A
very high osmotic driving forces [16]. Upon moderate heating polyester mesh is embedded within the polymer material for mechanical support.
(near 60 ◦ C) [17], the ammonium draw solutes decompose into The membrane thickness is less than 50 ␮m.
ammonia and carbon dioxide gases that can be separated by stan-
dard means. The separated gases can then be used to regenerate
hydrophilic cellulose-based polymer [13]. Fig. 1 shows a cross-
the draw solution. This draw solute separation step in the FO
sectional view of the membrane taken by SEM, indicating that
desalination process will not be addressed in this paper.
the thickness of the membrane is less than 50 ␮m. It is evident
The objective of this paper is to investigate the relationships
that the structure of the CA membrane is quite different from
among the draw solution concentration, feed water salinity, and
standard RO membranes. RO membranes typically consist of
permeate flux behavior in the ammonia–carbon dioxide FO pro-
a very thin active layer (less than 1 ␮m) supported by a much
cess. The ammonia–carbon dioxide draw solution concentration
thicker porous polymer supporting layer [18]. These layers are
is varied over a wide range (up to 6 M), processing feed waters
mechanically supported by a fabric support layer. Fig. 1 indi-
with salinities as high as 2 M sodium chloride (NaCl). On the
cates that the CA FO membrane lacks any thick fabric layer.
basis of these results, the mechanisms governing the permeate
An embedded polyester mesh provides the mechanical support
water flux behavior in the ammonia–carbon dioxide FO process
normally provided by the thick fabric layer [13].
are elucidated and discussed.

2. Materials and methods 2.3. Forward osmosis crossflow set-up

2.1. Ammonia–carbon dioxide draw solution and sodium
chloride feed solution
The draw solution is made by mixing ammonium bicarbonate
(NH4 HCO3 ) and ammonium hydroxide (NH4 OH) with deion-
ized water at proper proportions to produce a solution of the
desired concentration of ammonium salts [15,16]. Higher con-
centrations of ammonia and carbon dioxide salts require a higher
ratio of ammonia to carbon dioxide, which favors the formation
of ammonium carbamate, a highly soluble ammonium salt. The
ammonia to carbon dioxide molar ratios used ranged from 1.2
for the 1.1 M draw solution to 1.4 for the 6 M draw solution.
The feed solution was composed of NaCl dissolved in deion-
ized water. Concentrations ranged from 0.05 to 2 M. The 2 M
NaCl concentration is indicative of 75% recovery of a 0.5 M
NaCl solution, which is close to an initial seawater concentra-
tion. The temperature of the feed solution was 50 ◦ C.
2.2. Forward osmosis membrane
The membrane tested in this work was provided by Hydration
Technologies Inc. (Albany, OR). The proprietary FO membrane,
denoted as CA in this paper, is thought to be made from a
Fig. 2 depicts the apparatus employed in our laboratory-scale
FO experiments. The crossflow membrane unit is a SEPA cell
(GE Osmonics, Trevose, PA), modified to have channels on both
sides of the membrane. The ammonia–carbon dioxide draw solu-
tion is flowing on the permeate side and the NaCl solution on
the feed (active layer) side. Co-current flow is used to reduce
strain on the suspended membrane. Mesh spacers are inserted
within both channels to improve support. The spacers also pro-
mote turbulence and mass transport. Variable speed peristaltic
pumps (Manostat, Barrington, IL) are used to pump the liquids.
A constant temperature water bath (Neslab, Newington, NH)
is used to maintain the feed solution temperature, whereas the
draw solution temperature is kept constant by a heating mantle
(Glas-Col, Terre Haute, IN). Both the feed and draw solutions
were held at the same temperature (50 ◦ C) during the FO tests
and were controlled to within ±1 ◦ C.
2.4. Water flux and salt rejection measurements
Water flux is determined by measuring the weight change
(Ohaus, Pine Brook, NJ) of the draw solution over a selected
time period at the initial stage of the process. As water transports
by osmosis across the membrane from the saline feed water
116 J.R. McCutcheon et al. / Journal of Membrane Science 278 (2006) 114–123
Fig. 2. Schematic diagram of the laboratory-scale forward osmosis system. Co-current crossflow of the feed NaCl and draw solutions is used. A scale connected to
a PC measures the mass of water permeating into the draw solution, from which permeate water flux is calculated.
into the draw solution, the draw solution weight increases. The
rate at which the volume increases (determined from the weight
change) divided by the membrane area yields the water flux.
For determining NaCl rejection, a sample of draw solution
is taken after a complete FO run and is boiled until all water
evaporates. During the boiling process, the dissolved ammo-
nium salts decompose into NH3 and CO2 gases. The remaining
solid NaCl is hydrated with a known volume of deionized water,
and the NaCl concentration is measured by a chloride selective
electrode (Cole-Parmer, Vernon Hills, IL). Based on the amount
of water passed into the draw solution over the course of the
experiment and the final amount of NaCl in the draw solution,
the permeate NaCl concentration is determined. The percent salt
rejection, R, is then calculated from


Cp
R = 100 1 − (1)
CF
where Cp and CF are the permeate and feed NaCl concentrations,
respectively.
2.5. Dependence of membrane water permeability on feed
NaCl concentration
Osmotic permeability measurements to determine weather
the water permeability coefficient changes at high feed NaCl
concentrations were taken with a redesigned FO cell having
well-controlled hydrodynamics. The crossflow cell is of plate-
and-frame design with a rectangular channel on each side of
the membrane. The channel has dimensions of 7.7 cm length,
2.6 cm width, and 0.3 cm height, providing an effective mem-
brane area of 20.0 cm2 . The crossflow velocity during the testing
was 21.4 cm/s. An RO cell with similar dimensions was used to
determine the pure water hydraulic permeability for comparison.
To conduct the osmotic permeability measurements, NaCl
solution was used on the active layer side of the membrane,
while deionized water ran along the backing (support) layer side.
Osmotic flux data were collected for NaCl concentrations rang-
ing from 0.05 to 2 M, corresponding to the feed solutions that
were used in the FO experiments described above. The tem-
perature of both solutions was maintained at 50 ◦ C, which was
the temperature used for the FO and pure water permeability
experiments.
High driving forces and correspondingly high water fluxes
can cause severe concentration polarization effects, which occur
on the permeate side of the membrane (in this case, the active
layer). To account for this phenomenon, the extent of concentra-
tion polarization was calculated from film theory. The Sherwood
number, Sh, was first determined by using either the laminar or
turbulent flow correlation for a rectangular channel [19,20]:


dh 0.33
Sh = 1.85 ReSc (2)
L
Sh = 0.04Re0.75 Sc0.33 (3)
where Re is the Reynolds number, Sc the Schmidt number, dh
the hydraulic diameter, and L is the length of the channel. By
 J.R. McCutcheon et al. / Journal of Membrane Science 278 (2006) 114–123
using tabulated diffusion coefficient, D, data for NaCl at 50 ◦ C
[21], the mass transfer coefficient, k, could be calculated from
kdh
Sh = (4)
D
Once we determine the mass transfer coefficient, k, the CP
modulus at each permeate flux, J, could be calculated using


Cm J
= exp − (5)
Cb K
The flux, J, in this equation is negative because the water flux is
in the direction away from the membrane active layer surface. In
other words, the CP effect in our case is dilutive, meaning that
the membrane surface concentration, Cm , will be lower than the
bulk concentration, Cb .
To determine the CP modulus in the transition flow regime,
we use the average between the values calculated with Eq. (2) for
laminar flow and Eq. (3) for turbulent flow [22]. We have used
this CP modulus for our calculations of the dilutive CP effects
as the Reynolds number for the system based on the channel
dimensions and flow rate is 2100.
3. Results and discussion
3.1. Water flux with variable feed and draw solution
concentrations
Water flux across a semi-permeable membrane when osmotic
pressure is the driving force is given by [19,23]:
J = Aσ π (6)
Here, A is the water permeability coefficient, σ the reflection
coefficient, and π is the difference in osmotic pressures across
the membrane between the draw and feed solution at the separat-
ing interface (i.e., at the membrane active layer). The reflection
coefficient approaches 1.0 for a completely rejecting membrane.
Because the rejection in most of our FO experiments was above
95% (to be discussed later in the paper), the reflection coefficient
for the FO runs presented in this paper was assumed to be unity.
It should be noted, however, that determination of the reflection
coefficient for our system involving ammonia–carbon dioxide
as a draw solution and NaCl as a feed solution may be com-
plicated by the leakage of free ammonia to the feed size of the
membrane. Because our experiments involved numerous com-
binations of draw and feed solution concentrations, the determi-
nation of the reflection coefficient for each individual run was
not feasible.
The bulk osmotic pressures of our draw (NH3 /CO2 ) and feed
(NaCl) solutions were calculated using software from OLI Sys-
tems Inc. (Morris Plains, NJ) and Aspen HYSYS® (Cambridge,
MA). The OLI Systems software uses thermodynamic modeling
based on published experimental data to predict the properties
of solutions over a wide range of concentrations and tempera-
tures. The molar concentrations of the added ammonium solutes
– ammonium bicarbonate (NH4 HCO3 ) and ammonium hydrox-
ide (NH4 OH) – for each draw solution were used to calculate the
osmotic pressure. The OLI software was also used to calculate
117
Fig. 3. Flux data for a variety of feed solution (NaCl) concentrations. The water
flux is presented as a function of the difference in bulk osmotic pressures of the
draw and feed solutions. Experimental conditions: crossflow velocity and tem-
perature of both feed and draw solution of 30 cm/s and 50 ◦ C, respectively.
Note that a water flux of 10 ␮m/s corresponds to 21.2 gal ft−2 d−1 (gfd) or
36.0 l m−2 h−1 .
the speciation and pH of the draw solutions. The pH of the draw
solutions ranged from 7.73 for the 1.1 M draw solution to 8.26
for the 6 M draw solution.
Our previous work on the ammonia–carbon dioxide FO
desalination process [15] discussed limited water flux results
for a single feed solution concentration. In this investigation we
examine the permeate water flux behavior over a wide range of
feed as well as draw solution concentrations. Fig. 3 presents per-
meate water flux data for different feed NaCl concentrations as
a function of the osmotic pressure difference, πD − πF , where
πD and πF are the bulk osmotic pressures of the draw and feed
solutions, respectively.
Assuming the water permeability coefficient, A, is constant in
all tests and σ ≈ 1, Eq. (6) would predict a proportional increase
in flux with increasing osmotic driving force, regardless of feed
solution concentration. Fig. 3, however, suggests that feed solu-
tion concentration may be a key parameter that influences water
flux in the FO process. The data in Fig. 3 indicate that for
similar differences in osmotic pressure between the bulk feed
and draw solutions (πD − πF ), higher concentration feed solu-
tions produce less flux. We must consider, however, that higher
concentration feed solutions require more concentrated draw
solutions to achieve the same theoretical osmotic driving force.
This phenomenon, therefore, may not be a feed salinity effect.
The draw solution may also play a role.
Assuming that the full osmotic driving force, referred to as a
theoretical driving force, can be realized, a theoretical water flux
can be calculated based on Eq. (6). We use a constant water per-
meability coefficient value of 5.69 × 10−12 m Pa−1 s−1 in these
calculations, determined by reverse osmosis tests of the CA
membrane at 50 ◦ C. Later in this paper, we will reassess this
118 J.R. McCutcheon et al. / Journal of Membrane Science 278 (2006) 114–123

Table 1
Water flux, performance ratio, and NaCl rejection results for the FO runs carried out in this study under the stated feed and draw solution concentrations (and the
corresponding osmotic pressures)
Feed πF (atm) Draw πD (atm) π (atm) Theoretical flux in Experimental flux in Performance Rejection
concentration concentration ␮m/s (or gfd) ␮m/s (or gfd) ratio (%)
(M) (M)

0.05 2.5 1.1 47.4 44.9 25.8 (57.9) 5.3 (11.2) 0.204 97.90
0.05 2.5 3 127.6 125.1 72.0 (152.9) 8.6 (18.3) 0.120 99.45
0.05 2.5 6 249.5 236.2 136.0 (288.6) 10.1 (21.4) 0.071 98.77
0.2 9.8 1.1 47.4 37.6 21.6 (45.91) 4.1 (8.7) 0.188 93.72
0.2 9.8 3 127.6 117.8 67.8 (144.0) 6.8 (14.5) 0.101 97.34
0.2 9.8 6 249.5 228.8 131.7 (279.7) 9.4 (19.9) 0.068 98.12
0.5 24.8 1.1 47.4 22.6 13.0 (27.6) 0.9 (2.0) 0.074 86.53
0.5 24.8 1.6 73.3 48.4 27.9 (59.2) 2.8 (6.0) 0.102 96.03
0.5 24.8 2 94.5 69.7 40.1 (85.1) 3.1 (6.6) 0.078 96.40
0.5 24.8 3 127.6 102.8 59.2 (125.6) 4.6 (9.7) 0.077 98.07
0.5 24.8 4.5 193.3 168.5 97.0 (205.9) 5.9 (12.5) 0.061 97.14
0.5 24.8 6 249.5 213.8 123.1 (261.3) 6.4 (13.5) 0.049 98.70
1 51.5 3 127.6 76.1 43.8 (93.0) 2.3 (4.8) 0.052 93.89
1 51.5 4.5 193.3 141.8 81.6 (173.2) 4.4 (9.4) 0.054 95.92
2 113.8 4.5 193.3 79.5 45.8 (97.2) 1.1 (2.3) 0.024 90.14
2 113.8 6 249.5 124.9 71.9 (152.7) 2.0 (4.3) 0.026 96.85

assumption regarding the water permeability coefficient. It is
also assumed that σ ≈ 1 as discussed earlier in the paper. In
all experiments presented in Fig. 3, the measured water flux
was lower than the theoretical flux. We define the performance
ratio as the experimental water flux divided by the theoretical
water flux. This ratio is equivalent to the percentage of the bulk
osmotic pressure difference that is effectively generating water
flux across the FO membrane.
Table 1 summarizes the FO experimental data, including the
feed and draw solution concentrations used, the corresponding
osmotic pressures and the osmotic pressure difference (π), the
calculated (theoretical) water flux based on the osmotic pressure
difference, the measured water flux, and the performance ratio.
Also included in this table are salt rejection data, to be dis-
cussed later. The variation of the performance ratio with draw
solution concentration for the different feed solution concen-
trations is shown in Fig. 4. The results demonstrate that the
performance ratio decreases with increasing draw and feed solu-
tion concentrations. At best, around 20% of the osmotic pressure
driving force is utilized, while at higher draw solution concen-
trations, as little as 5% or less of the osmotic pressure driv-
ing force is realized for the experimental conditions discussed
here.
3.2.1. External concentration polarization
In pressure-driven membrane processes, convective perme-
ate flow causes a buildup of solute at the membrane active layer
surface. Referred to as CP, this reduces permeate water flux
due to increased osmotic pressure that must be overcome with
hydraulic pressure [24–26]. Permeate convection, and hence CP,
is not limited to pressure-driven membrane processes and also
occurs in FO on the membrane feed side. This may in part explain
the low performance ratios in Fig. 4. A similar but dilutive
effect occurs on the permeate (support layer) side, as permeat-
ing water dilutes the draw solution along the external membrane
backing layer surface. Higher concentration draw solutions will
increase the severity of this “dilutive” CP, since higher draw
solution concentrations yield higher water flux rates. Combined,
the “concentrative” and dilutive external CP phenomena reduce

3.2. Processes governing the permeate flux behavior

The much lower than expected water flux in the previously

described experiments is likely caused by several membrane
associated transport phenomena. Specifically, two types of con-
centration polarization (CP) phenomena – external CP and inter-
nal CP – can take place in FO as discussed below. In addition,
exposure of the membrane to very concentrated salt solutions
Fig. 4. The effect of draw solution concentration on the performance ratio for
such as those used in our FO runs may influence the water per- various feed solution concentrations. Experimental conditions: crossflow rate
meability of the membrane. These possible causes are analyzed and temperature of both feed and draw solution of 30 cm/s and 50 ◦ C, respec-
below. tively.
 J.R. McCutcheon et al. / Journal of Membrane Science 278 (2006) 114–123 119

the effective osmotic driving force. In a scaled up process, the
effects of the external CP phenomenon on driving force can
be minimized by using crossflow filtration with hydrodynam-
ics designed to produce sufficient shear and turbulence at the
membrane surface [27].
Our system (SEPA cell) incorporates a shallow channel
(0.864 mm) and mesh spacers, minimizing both concentrative
and dilutive external CP effects through high shear and increased
turbulence. For the experimental conditions of our FO runs
(crossflow velocity of 30 cm/s and a temperature of 50 ◦ C), the
concentration polarization modulus values were calculated from
film theory for the feed side of the membrane. Reynolds num-
bers in the system ranged from about 940 to 960 (the range is
due to viscosity and density variability over the range of NaCl
concentrations).
We calculated the mass transfer coefficients and the CP mod-
uli for each of our tests based on Eqs. (2), (4), and (5). Note that
for this case, Eq. (5) has a positive exponent. We did not use the
turbulent correlation because the flow in this case is in the lam-
inar regime. CP moduli for the various feed solutions and water
fluxes ranged from 1.03 to 1.33, meaning that effective mem-
brane surface concentration was not much greater than the bulk
concentration. Since the feed NaCl concentration was always
significantly less than the draw solution concentration, small
increases in the NaCl surface concentration at the active layer
did not critically affect overall driving force. It is also important
to note that the higher CP moduli were associated with the lower
feed NaCl concentrations (due to higher water fluxes), further
mitigating the effect of external CP on the driving force. Exter-
nal CP on the feed side of the membrane, therefore, can be ruled
out as the cause for the markedly lower than expected water flux.
Dilutive external CP (i.e., on the outer support layer side) was
not considered since the water permeating the backing layer of
the membrane contained some concentration of the draw solute.
This permeate concentration would mitigate the polarization
effects on the backing side of the membrane since pure water is
not entering the bulk solution at the membrane surface. Polar-
ization effects on the permeate side of the active layer within the
porous support layer are considered in the next section.
3.2.2. Internal concentration polarization
When an osmotic pressure gradient is established across a
completely rejecting dense symmetric membrane, as depicted in
Fig. 5a, the driving force is the difference in osmotic pressures
of the bulk solutions in the absence of concentrative and dilutive
external CP. When a composite or asymmetric membrane con-
sisting of a dense separating layer and a porous support layer is
used in forward osmosis, two phenomena can occur depending
on the membrane orientation.
If the porous backing layer of an asymmetric membrane is
facing the feed solution, a polarized layer is established along
the inside of the dense active layer as water and solute propagate
the porous support layer. Referred to as internal concentration
polarization [3,4], this phenomenon, illustrated in Fig. 5b, is

Fig. 5. Illustration of driving force profiles, expressed as water chemical potential, µW , for osmosis through several membrane types and orientations. (a) A
symmetric dense membrane. (b) An asymmetric membrane with the porous support layer against the feed solution; the profile illustrates concentrative internal CP.
(c) An asymmetric membrane with the dense active layer against the feed solution; the profile illustrates dilutive internal CP. The actual (effective) driving force is
represented by µW . External CP effects on the driving force are assumed to be negligible in this figure.
120 J.R. McCutcheon et al. / Journal of Membrane Science 278 (2006) 114–123

similar to concentrative external CP, except that it takes place

within the protected confines of the porous backing layer, and
hence cannot be minimized by crossflow.
Our system, however, is run with the membrane in the oppo-
site orientation. In our FO configuration, the NaCl feed is
channeled on the active layer side of the membrane and the
ammonia–carbon dioxide draw solution is channeled on the
backing layer side. This was done to most closely mimic the
typical membrane desalination configuration where the feed is
directed against the active layer. Upon contact with the back-
ing layer of the membrane, the draw solution will diffuse into
the porous substructure. The osmotic driving force between the
feed and draw solutions will then be established across the dense
active layer of the membrane. As water permeates the active
layer, the draw solution within the porous substructure becomes
diluted. We refer to this as dilutive internal CP. As with con-
centrative internal CP, dilutive internal CP cannot be controlled
by crossflow. The unmitigated dilutive internal CP phenomenon
significantly decreases the net driving force, leading to decreased
flux. Illustrated in Fig. 5c, this phenomenon occurs when asym-
metric or thin film composite membranes are used in an FO mode
with the draw solution against the porous support/backing layer.
For dilutive internal CP to occur in our tests, the CA mem-
brane would have to contain a porous support layer. Fig. 1 does
not reveal any obvious asymmetry within the CA membrane.
If there were no porous support layer, then no dilutive internal
CP would occur. However, a closer inspection of the membrane,
shown in Fig. 6, uncovers asymmetry. Fig. 6a shows a dense
active layer, while Fig. 6b reveals a porous supporting structure.
This porous structure should result in substantial dilutive inter-
nal CP that explains the much lower than expected water flux. Fig. 6. SEM images of (a) the FO membrane (CA) active layer (at 100,000×)
Additional analysis of the marked effect of dilutive internal CP and (b) the membrane backing layer (100,000×). The image of the membrane
in our system is given later in this paper. backing layer reveals a porous structure.

3.2.3. Does the membrane water permeability coefficient
change with feed NaCl concentration?
Knowing the membrane water permeability at a given set of
system conditions is critical to calculating the theoretical flux
and performance ratio as well as eliminating uncertainty in the
cause of various flux behaviors. Previous studies on osmosis
through asymmetric cellulose acetate membranes have reported
that membrane water permeability is often found to be signif-
icantly lower than the corresponding hydraulic water perme-
ability determined by RO [28–32]. Pusch [32] presented data
showing a precipitous drop in water permeability as the mem-
brane became exposed to more salt. Much of the data, however,
was in the form of observations of apparent permeability based
on water flux and bulk solution driving force rather than the true
membrane permeability. This behavior was attributed to inter-
nal concentration polarization effects within the porous support
layer. A few studies, however, suggested that high salt concentra-
tions would actually lower the intrinsic water permeability of a
membrane through osmotic deswelling or other related phenom-
ena [29,30]. Our performance ratio data was calculated using the
hydraulic water permeability coefficient obtained for pure water
with RO at 50 ◦ C. In order to test the influence of NaCl feed
solution on the water permeability coefficient, we carried out
osmotic permeability runs exposing the active layer to increas-
ing NaCl feed concentrations (up to 2 M) while DI water was
placed against the support layer.
Fig. 7 shows the flux data gathered, based on the experimental
procedure outlined in Section 2.5. As the NaCl bulk solution con-
centration is increased, the flux behavior is not linear. This may
suggest that the permeability of the membrane may be chang-
ing at higher NaCl concentrations. The water fluxes, however,
are quite high for the tests at high NaCl concentrations where
nonlinearity is obvious, suggesting that dilutive external CP on
the permeate side of the membrane may significantly reduce the
osmotic driving force.
To test our assumption that dilutive external CP rather than
reduction in membrane water permeability is the cause for the
deviation from linearity, we carried out CP calculations as out-
lined in Section 2.5. The Reynolds number was found to be
around 2100 for the flow conditions in the new FO cell described
above. With a Reynolds number on the cusp of the laminar flow
regime, coupled with the mesh spacers, transition flow is likely
occurring. We therefore used Eqs. (2)–(5) to account for this
phenomenon and calculated the NaCl concentration at the mem-
brane surface using the average of the turbulent and laminar CP
modulus for each test [22]. The effective osmotic driving force is
 J.R. McCutcheon et al. / Journal of Membrane Science 278 (2006) 114–123
Fig. 7. Flux data with NaCl draw solution on the active layer and deionized water
on the backing layer. Experiments were carried out with a specially designed
FO cell described in Section 2.5. Experimental conditions: crossflow velocity
and temperature of both feed and draw solution of 21.4 cm/s and 50 ◦ C, respec-
tively. Note that a water flux of 10 ␮m/s corresponds to 21.2 gal ft−2 d−1 (gfd)
or 36.0 l m−2 h−1 .
then calculated using the OLI software as before. This corrected
data is presented in Fig. 8.
Fig. 8 shows that flux increases linearly with increasing
osmotic pressure at the membrane surface when the concen-
tration at the membrane surface is calculated using film theory.
Also plotted on Fig. 8 is the hydraulic permeability data taken
for the CA membrane using RO. The osmotic flux data matches
closely to the hydraulic flux data, allowing us to conclude that
the water permeability coefficient of our CA membrane does not
change when exposed to the NaCl feed concentrations used in
the FO runs (summarized in Table 1). This confirms that changes
in membrane permeability are not affecting water flux.
Fig. 8. Dependence of water flux (of data in Fig. 7) on the effective driving
force after correction for external dilutive concentration polarization along the
active layer. In these experiments, water transport is from the DI water feed
(on the support layer side) to the NaCl solution (on the active layer side). The
effective osmotic driving force is calculated based on the NaCl concentration
at the membrane surface (active layer side) after corrections for dilution due to
external CP. The hydraulic driving force data was collected in an RO cell with
pure water and the same FO membrane. Experimental conditions for osmosis
data: crossflow rate of 21.4 cm/s and temperatures of 50 ◦ C. RO data conducted
with an RO cell of identical dimensions. Note that a water flux of 10 ␮m/s
corresponds to 21.2 gal ft−2 d−1 (gfd) or 36.0 l m−2 h−1 .
121
Fig. 9. Flux data (for the FO experiments summarized in Table 1) plotted against
the logarithm of the ratio of feed and draw solution osmotic pressures. Experi-
mental conditions: SEPA cell with crossflow rate and temperature of both feed
and draw solution of 30 cm/s and 50 ◦ C, respectively. Note that a water flux of
10 ␮m/s corresponds to 21.2 gal ft−2 d−1 (gfd) or 36.0 l m−2 h−1 .
3.3. Water flux uniquely depends on a normalized osmotic
pressure driving force
Another way of evaluating the flux behavior and confirming
the presence of dilutive internal CP is by normalizing the driv-
ing force. Two methods are presented here. The first, suggested
by Loeb et al. [5], considers the ratio of draw solution to feed
solution bulk osmotic pressure, πD /πF . It was shown that the
effective driving force in a forward osmotic process that is influ-
enced by internal CP can be approximated by ln(πD /πF ). The
data from Table 1 are normalized using this method and shown
in Fig. 9. The linear trend confirms that our data correspond
well with such analysis of flux behavior in an osmotic process
through an asymmetric RO membrane.
To better conceptualize how the concentration of draw solu-
tion affects dilutive internal CP, we use another normalization
technique. By plotting our water flux against the bulk osmotic
pressure difference normalized by the feed osmotic pressure
(πD − πF )/πF , a strong logarithmic trend is apparent (Fig. 10).
This observation supports our conclusion that the water flux
behavior in governed by dilutive internal CP. At low normalized
driving forces, flux increases sharply with increasing driving
force. However, at higher driving forces caused by higher draw
solution concentrations, flux increases diminish as driving force
increases. High draw solution concentrations generate larger
osmotic driving forces which produce more water flux. Higher
water fluxes increase the severity of dilutive internal CP by
increasing the degree of dilution within the backing layer of
the membrane, lowering the performance ratio and diminishing
the increase in flux for higher draw solution concentrations.
3.4. Salt rejection and recovery
Even with dilutive internal CP, water flux is still reason-
ably high. For an efficient FO desalination process, it is also
imperative that salt rejection be high. Rejection data for all our
122 J.R. McCutcheon et al. / Journal of Membrane Science 278 (2006) 114–123
Fig. 10. Flux data (for the FO experiments summarized in Table 1) plotted
against the normalized osmotic pressure difference. Note that πD is the bulk
osmotic pressure of the draw solution and πF is the osmotic pressure of the feed
solution. Experimental conditions: SEPA cell with crossflow rate and tempera-
ture of both feed and draw solution of 30 cm/s and 50 ◦ C, respectively. Note that
a water flux of 10 ␮m/s corresponds to 21.2 gal ft−2 d−1 (gfd) or 36.0 l m−2 h−1 .
Fig. 11. Rejection data for the FO runs presented earlier (Table 1). Experimental
conditions: SEPA cell with crossflow rate and temperature of both feed and draw
solution of 30 cm/s and 50 ◦ C, respectively. Note that a water flux of 10 ␮m/s
corresponds to 21.2 gal ft−2 d−1 (gfd) or 36.0 l m−2 h−1 .
experiments summarized in Table 1 are presented in Fig. 11. The
results indicate that NaCl rejection is above 95% for most of the
experiments, and approaches 99% for some of the higher water
flux experiments. Rejection increases with increasing water flux
due to the “dilution effect”, which occurs when water flux is
increased without a subsequent increase in salt flux. Increasing
the osmotic pressure driving force only affects water flux and
not salt flux, and rejection is thereby improved.
4. Concluding remarks
The data presented in this paper demonstrate that the draw
solution concentration has a significant effect on the flux behav-
ior of the ammonia–carbon dioxide FO desalination process.
It was determined that dilutive internal concentration polariza-
tion is primarily responsible for a much lower than expected
water flux. Data showed that increasing the draw solution con-
centration leads to a decrease in the percentage utilization of
the available bulk osmotic driving force, or a decrease in per-
formance ratio, due to an increase in the severity of the dilutive
internal CP. This was concluded after external CP and possible
osmotic deswelling of the FO membrane were shown to have
negligible effects on FO performance. The performance data
show that in spite of dilutive internal CP, water flux is still rea-
sonably high with the FO membrane. The FO system was shown
to perform well in terms of NaCl rejection and to operate at very
high feed concentrations, simulating operation at high feedwa-
ter recovery. While overall performance of the FO process was
satisfactory, the FO membrane tested in this study is not opti-
mal for this process since its primary use is not for desalination.
The development of a semi-permeable membrane having high
salt rejection and minimal internal CP will lead to improved
performance of this forward osmosis desalination process.
Acknowledgements
The authors acknowledge the support of the Office of
Naval Research, Award No. N000140311004. The authors also
acknowledge Hydration Technologies for providing membranes
for this research.
References
[1] B. Van der Bruggen, L. Lejon, C. Vandecasteele, Reuse, treatment, and
discharge of the concentrate of pressure-driven membrane processes,
Environ. Sci. Technol. 37 (2003) 3733–3738.
[2] T.Y. Cath, V.D. Adams, A.E. Childress, Experimental study of desalina-
tion using direct contact membrane distillation: a new approach to flux
enhancement, J. Membr. Sci. 228 (2004) 5–16.
[3] G.D. Mehta, S. Loeb, Internal polarization in the porous substructure of
a semipermeable membrane under pressure-retarded osmosis, J. Membr.
Sci. 4 (1978) 261–265.
[4] K.L. Lee, R.W. Baker, H.K. Lonsdale, Membranes for power-generation
by pressure-retarded osmosis, J. Membr. Sci. 8 (1981) 141–171.
[5] S. Loeb, L. Titelman, E. Korngold, J. Freiman, Effect of porous support
fabric on osmosis through a Loeb-Sourirajan type asymmetric mem-
brane, J. Membr. Sci. 129 (1997) 243–249.
[6] G.W. Batchelder, Process for the demineralization of water, US Patent
3,171,799 (1965).
[7] D.N. Glew, Process for liquid recovery and solution concentration, US
Patent 3,216,930 (1965).
[8] B.S. Frank, Desalination of sea water, US Patent 3,670,897 (1972).
[9] W.T. Hough, Process for extracting solvent from a solution, US Patent
3,532,621 (1970).
[10] W.T. Hough, Solvent extraction into a comestible solute, US Patent
3,702,820 (1972).
[11] J. Yaeli, Method and apparatus for processing liquid solutions of suspen-
sions particularly useful in the desalination of saline water, US Patent
5,098,575 (1992).
[12] K. Stache, Apparatus for transforming sea water, brackish water, polluted
water or the like into a nutritious drink by means of osmosis, US Patent
4,879,030 (1989).
[13] K. Lampi, Forward osmosis pressurized device and process for gener-
ating potable water, US Patent 6,849,184 B1 (2005).
[14] R.A. Neff, Solvent extractor, US Patent 3,130,156 (1964).
 J.R. McCutcheon et al. / Journal of Membrane Science 278 (2006) 114–123
[15] J.R. McCutcheon, R.L. McGinnis, M. Elimelech, A novel ammonia-
carbon dioxide forward (direct) osmosis desalination process, Desalina-
tion 174 (2005) 1–11.
[16] R.L. McGinnis, Osmotic desalination process, US Patent Pending
PCT/US02/02740.
[17] M. Trypuc, U. Kielkowska, Solubility in the NH4 HCO3 + NaHCO3 +
H2 O system, J. Chem. Eng. Data 43 (1998) 201–204.
[18] R.W. Baker, Membrane Technology and Applications, Wiley, West Sus-
sex, 2004.
[19] M. Mulder, Basic Principles of Membrane Technology, Kluwer Aca-
demic, Dordrecht, 1996.
[20] V. Gekas, B. Hallstrom, Mass-transfer in the membrane concentration
polarization layer under turbulent cross flow. I. Critical literature-review
and adaptation of existing sherwood correlations to membrane opera-
tions, J. Membr. Sci. 30 (1987) 153–170.
[21] V. Lobo, Mutual diffusion-coefficients in aqueous-electrolyte solutions
(technical report), Pure Appl. Chem. 65 (1993) 2614–2640.
[22] Y.A. Le Gouellec, M. Elimelech, Control of calcium sulfate (gypsum)
scale in nanofiltration of saline agricultural drainage water, Environ.
Eng. Sci. 19 (2002) 387–397.
[23] W. Pusch, R. Riley, Relation between salt rejection-r and reflection
coefficient-sigma of asymmetric cellulose-acetate membranes, Desali-
nation 14 (1974) 389–393.
[24] L.F. Song, M. Elimelech, Theory of concentration polarization in cross-
flow filtration, J. Chem. Soc. Faraday Trans. 91 (1995) 3389–3398.
View publication stats
123
[25] S.S. Sablani, M.F.A. Goosen, R. Al-Belushi, M. Wilf, Concentration
polarization in ultrafiltration and reverse osmosis: a critical review,
Desalination 141 (2001) 269–289.
[26] M. Elimelech, S. Bhattacharjee, A novel approach for modeling con-
centration polarization in crossflow membrane filtration based on the
equivalence of osmotic pressure model and filtration theory, J. Membr.
Sci. 145 (1998) 223–241.
[27] E.M.V. Hoek, A.S. Kim, M. Elimelech, Influence of crossflow membrane
filter geometry and shear rate on colloidal fouling in reverse osmosis
and nanofiltration separations, Environ. Eng. Sci. 19 (2002) 357–372.
[28] I. Goosens, A. Vanhaute, Use of direct osmosis tests as complementary
experiments to determine the water and salt permeabilities of reinforced
cellulose-acetate membranes, Desalination 26 (1978) 299–308.
[29] G.D. Mehta, S. Loeb, Performance of permase B-9 and B-10 membranes
in various osmotic regions and at high osmotic pressures, J. Membr. Sci.
4 (1979) 335–349.
[30] H.A. Massaldi, C.H. Borzi, Non-ideal phenomena in osmotic flow
through selective membranes, J. Membr. Sci. 12 (1982) 87–99.
[31] G.D. Mehta, Further results on the performance of present-day osmotic
membranes in various osmotic regions, J. Membr. Sci. 10 (1982) 3–19.
[32] W. Pusch, Transport behaviour of asymmetric cellulose acetate mem-
branes in dialysis and hyperfiltration, in: H.B. Hopfenberg (Ed.), Per-
meability of Plastic Films and Coatings to Gases, Vapors and Liquids,
American Chemical Society, Division of Organic Coatings and Plastics
Chemistry.