Desali~tation, I8 (1976) 297-306

@ Elsevicr %cntific Publishing Company. Amsterdam - Prantcd in The S&crlan&

DRINKING WATER FROM SEA WATER BY FORWARD OSMOSlS

J. 0. KESSLEK* AVV C. D. MOODY t

l Pi1.bsits Depurtttt~*rtf and t&hoot of Rerte~ crhlc Rerorrt ces. Cstiwrsiry of Arizorto. Tttcron, Ariz.
35721 (U.S.A.I
(Rccciux! July 7. 1916)

SUU.\lARY

Potable water may be extracted from a saline source, such as the ocean, using
a conccntratcd solution of nutrients and a scmipcrmeable membrane. This paper
applies the previously presented theory (I) of the forward osmosis es;lr;Lclor 10 rhis
case and compares that theory with experiment. Theory and experiment are found
to match well. Calculated and observed magnitudes of water extraction rnfc and
nutrient utihzation indicate that practical devices may be constructed. It is found
that the volume of waler obtained per mass of nutrient meets the human rcquire-
ment for fluids and nutrient. Optimization of Ihe system is discussed.

SY hl BOIS

cm - molal cont. of nutrients in driving solution at location s

(-JO) - feed cont. of nutrients
CJl’) - diluted product cont. of nutrients in final dricking fluid
=5 - molal concentration of the source solution (sea water, assumed = 0.6
NaCl in this paper)
c,,J/t) - molal cont. of source salt in tinal drinking fiu~d
c;,(A) - average salt cont. of membrane fluu ($(/I) = c&)Q~(/z)[QJII) -

cp(s) - salt cont. of the focal membrane flux at location s

tf - subscript denoting “driving” or “nutrient” solution
11 - length of the extractor membrane in cm (//I = total membrane area)
I - width of the extractor membrane in cm (f = circumference for tubular
membranes)
L - mcmbranc water permeability coeflicient in cm/s atoll
qJ_u) - local membrane water flux in (cm’/cm’ s) = cm/s
q,(lr) - average membrane water flux for the extractor in cm/s, (I,,,(/,)= Q,(/~)/plir
QJO) - feed rate of water in the driving solution in g/s
QM - rate at which water leaves the extractor in g/s (Q,(A) = Qd(0) +
Q,(~J)) ’
298 3. 0. KfSLER ASD C. D. MOODY

QmW - total membrane water flux in g/s; Q,,,(/r) = p/j<:q,,,(x)d-r

Q,(/I) - feed rate of water in the source solution in g/s
R - universal gas constant = 0.0521 liter atm/mole/de_eree
\ - subscript denotin_e “source” or “salty” solution
SRSo - reverse osmosis solute rejection at a total difference of 50 atm
T - absolute temperature (‘K)
IOi - membrane permeability coeff. for solute i in cm/s
-Ii - osmotic coeficicnt of solute i in :he van ‘t Hoff relation
s - boundary layer thickness (cm)
/( - absolute viscosity (g-s- ‘-cm-‘)
xi - osmotic pressure in atmosphere, due to solute i_ zi = r,ciRT
P - density of water _c:cm’. taken as unity in this paper

INTRDDUC-I-IOS

One of the most interesting situations where the forward osmosis solvent
extraction technique may be applied is in the derivation of emergency potable
water supplies in situations where there is little storage capacity available. as in
lifeboats or other small craft. The process can be used to tmnsfer water from the
sea into a concentrated nutrient solution (2) placed in a forward osmosis extractor.
The weight of the only stored consumable required. the soluble foodused to extract
water. is expected to be less than one sixth of the weight of the normally carried
food plus water.
The theory of forward osmosis extraction has been discussed in a previous
paper (I). The mathematical models presented there are used in this paper to
furnish quantitative predictions relating to the transfer of fresh water from an
appropriate infinite supply of sea water into a concentrated solution of appropriate
nutrients such as various low molecular weight sugars and essential amino acids.
The model predictions are then compared uith experimental results. Another
objective of this paper is to develop insights into the choice of physical components
which ivilt optimize the water extraction process, i.e., which will yield the most
potable water, for the smallest amount of solute, the smallest extraction apparatus,
and at the highest transfer rate.

ALLOWABLE SALT COXI-ENT IN DRINKISG WATER

No practical perfectly salt rejecting membranes exist. Also, for a given

membrane type and fabrication technique, the membrane water permeability
coethcient increases with salt permeability, i.e., poor salt rejection correlates with
high water transfer rate. It is therefore necessary before proceeding further to
state the amount of salt that can be assumed tolerable in an efficient potable water
supply.
DRINKING WATER FROM SEA WATER BY FORWARD tKMOSIS 299

TABLE 1

0.0 2-l 4.8

0.1 2.0 4.9
0.15 2-7 5.1
0.2 2.8 5.6
0.25 3.1 6.5

The minimal daily requirement of drinking fluids is &en in Table I (3).

It is seen that at 90°F and a salt content beyond 0. i7 osmolal the volume trquire-
ment begins fo increase steeply. This concentration will therefore be taken as a
design guide. The rather substantial amount of metabolic water from the nutrients
will be ignored for the sake of conservatism.

DESCRIPTKIN OF THE POR\VARD aMOSIS EXTRACTOR

A schematic of a forward osmosis extractor is shown in Fig. 1. A differentially

permeable mcmbranc, either flat or as a tube, separates the nutrient driving
solution from the sea water source solution. The nutrient solution enters at the
inlet with concentration c,,(O) and solbent feed rate Qd(0). The sea water enters with
concentration c,(/r) = cS(c, = 0.6 molai NaCi in this paper) and solvent feed rate
Q,(h). An intinite supply of sea water is assumed. Q&it) is set much larger than the
membrane flux so that the concentration of the seawater leaving the extractor at
s = 0 is essentially eS.
Water flows through the membrane from the sea water into the nutrient
solution due fo the higher osmotic pressure of rhc nurrient solution. Osmotic
pressure is proportional to concentration by the van ‘t Hoff relation. zri .= rr,c,Rf.

INLET
~CONCENTRATEO
I OUTLET FOR
DRINKING FWIO
Y’ ORWING SOLUTION
NUTRIENT fDlLUTE NUTRIENT
SCWJflON~ + +/tg,~j --MEUERANE S”‘uffoM
J-t- - c
QUTLEf SOURCE SOLUTION
fSEAWATERf 7 ? f-- I%%ATERl
..~
XJO -1 x=h

Fig. 1. Schcmattc of a forw~& osmosis countercurrent extractor. For extracting fresk

water from ~3 water the sowcc solution is sea water and the driving solution contains human
nutrients.
300 J. 0. KESSLER AND C. 0. MOODY

As the nutrient solution proceeds downstream and its voiume is augmented

by incoming membrane water, c, decreases ar.d so does the local membrane water
flow rate, q,,,(x). The product d.qinking fluid emerges with a concentration &I)
much less than cd(O) and with a water flow rate QJ(/r) equal to Q,(O) plus the water
extrac:ed through the membrane_ Q,,,(/I).

OPTIMUM MEMBRASE SELECTIOS

Generally the membrane is slightly permeable to the solutes so that sea salt
passes through the membrane into the nutrient solution. The salt concentration in
the nutrient solution is denoted by c,,. As explained in a previous section, we shall
rcquirc

C&/I) I 0. I%, (= 0.17 osmolal electrolytes)

in the produced drinking fluid. In addition the extractor must be operated so that
everywhere the salt concentration of the membrane permeate, C,(X), is less than
c,. Other factors to consider are the loss of nutrients through the membrane and
the physical size of the extractor.
There is a wide range of membrane transpo:t properties to consider in the
selection of the optimum membrane, one which will produce the highest transfer
rate of water with the lowest transfer rates of sea salt and nutrient solute. For a
given membrane type the permeabilities to water, sodium chloride and nutrients
(L, ~~~~~~and wJ* vary simultaneously. Using data from Sourirajan (e for
cellulose acetate membranes which were initially treated at a hydrostatic pressure
of 20.4 atm (300 psi) ihc water and sodium chloride permeabilities vary according
to the empirical relation, A!. = 9.38 x IOMJ (w,,,,)‘*“~. There is fess data in the
literature for the cellulose acetate rejection of human nutrients such as glucose.

relationship. w~~~~,,JL =
0.24 (H&c,/L) o-42. Although the basis for the equation is limited, the two points

l The membrane transport properties (L, wd and ~*s~cl) arc defined to relate fluxes to the gndicnt
causing the tIux.
Solute Bux (moles s-1 cm-l) = w (cm 5-I) x (conanrration difference across mcm-
brane: moles cm-J)
Solvent tlu~ (cm3 s-1 cm-“) E L (cm s-1 atm-1) Y (effective pressure ditiercncc
across membrane, atm)
The relation, w;L = (1 - SR& :< 50 atm., was used to convert reve&e osmosis solute rejections
to values for w and L. SRJO is the reverse osmosis solute rejection at 50 atm. hydrostatic and
osmotic pressure difference across the membrane. The revetxe osmosis solute rejection is defined
by Mertens (7) as: SR = (cl - cp)/q 2( 1 - (wfL(Ap - AZ?)].
aJ - AT is the effective pressure difference at which SR is evaluated, and cf and c,, are the
concentrations of the feed and permeate solutions respectively.
DRlNKlNG WATER FROM SEA WATER BY FORWARD DSMDSiS 301

correspond to the region of interest, and the actual value of IV~,~_~ is not crucial
at this time so long as w*~,~_,%~ is small.
Assuming the permeability relations given, one may comptute the operating
characteristics of an extractor utilizing a cellulose acetate mem’tirane. Assuming
transfer of solutes and solvents. but ignoring polarizatior effectqMode1 2 of (I)j,
the calculation was performed using the Runge-Kutta Mcrson integration technique
contained in version 3 of DARE P (6).
Five different NaCI permcabilities were employed, the results for one case
are shown in Fig. 2. FRESH is the number of liters of water extracted per kilogram
of nutrient solute supplied. c:,(A) is the salt concentration in the drinking fluid
extracted. c,(h) is the salt concentration of the local membrane permeate. q is the
fraction of nutrient which remains on the driving side of the extractor. q,,,(/z) is the
average membrane water flux in cm/set. (To convert from cm/s to U.S. gallons
per day per square foot, multiply by 2.12 x IO’).
If ci,(/f)/c, is less than 0.15. then sea water can be added to the extracted
water until its concentration is 0. I3 c,. The total water thus obtained per kilogram
of nutrient can be calculated by the equation
Total waterjnutrient = FRESH * (1 - c~(/I)/c,)/(~ - .lS).
Table II lists the outputs which result from empioying the conditions:
&(/1)/c, _< 0.15, c,(/l),/c, 5 1.0 and total water//ltjtime 2 2.7 x lob4 g-cm-’ -s-l.
(The last condition is arbitrary at this point and is used solely for the purpose of
comparing the predicted results of membranes C, D and E).

FRESH/IO

lh/O,b) in (e )-
Fig. 2. Results predicted by Model 2 of a forward osmosis extractor for the extraction of
drinkin, m-eater
.+ from sea water. C&I) and c’ti th) are the local and total salt cones. of the membrane
flux; FRESH is liters of water extracted per kg nutrient; 111is the membrane area in cm’; (7,,,(h)
is the aterage membrane water tlux in g-cm%-‘: @n(O) is the nutrient solvent feed rate in g/s:
11is the nutrient fraction conserved. The membrane properties are: 15 = 3.05 x lob5 g-em-z-s-1-
arm-l, w’.\-~cI = 7.62 x 10-s~/s, wd = 4.0 x lO-G cm/s. The operating parameters are: cd(O) =
10.0 molal glucose-fructose, c, = 0.6 molal NaCI. ad = 1.04, a, = 1.86, RT = 24.5 liter-atm-
molt-1. Model 2 neglects concentration polarization (1).
302 J. 0. KESSLER AND C. 0. MOODY
DRINKISG WATER FROM SEA WATER BY bORWARD OSXlOSIS 303

it CZHIbe seen from Table II that membrane D produces the largest volume of
0. I7 osmolal water per mass of nutrient_ 5.2 liters per kg, based on the above three
output conditions and the assumption of no concentration polarization.

SURVtVAL AT SEA. SUMMAUY 01’ TIIEORY

For every kilogram of nutrient powder stored on a lifcraft ( I .5 kg of solution

if stored as a 641:; nutrient solution). approximately 5.2 liters of drinking water
containing 0. I7 osmolal electrolytes can be obtained from sea water containing
I. I osmolal electrolytes at a rate of 2.7 x iOmJ cm/s (5.7 gpd/ft’). 520 g of nutrient
and I I6 square cm of membrane (0.11 sq. ft.) can supply 2ooO food calories and
2.7 liters of drinking water, the daily fluid requirement at 80°F.
The assumption of no concentration polarization used above, overestimates
the membrane water flux. However, because of the absence of membrane compac-
tion, the vvater permeability coelhcient, L. can be expected to be larger for mem-
branes used in the forward osmosis mode than for membranes used in reverse
osmosis. given a particular sodium chloride permeability_ One may therefore
expect i. > 9.4 x IO-’ (tr’N.c, )“.19 for forward osmosis and higher water lluxes
than calculated above. These factors may well balance each other out.

tXl’tRlhlESfAL

In the experiment reported here the extractor shown schematically in Fis. I

is oriented with the flat membrane in a horizontal position. In order to provide
passive mixing by fret convention the glucose-fructose driving solution is above
the membrane, and the sodium chloride source solution is below the membrane.
The smooth, active side of the cellulose acetate membrane faces the driving
solution. The source solution concentration. c,, is kept constant along the length
of the estractor by maintaining the source solution feed rate large relative to the
membrane water flux. The output tube of the driving solution is higher than the
output of the source solution so that a slight hydrostatic pressure opposes the
membrane water flow. This hydrostatic pressure difference assures that the
membrane water flow is due only to osmotic Row and is not due to bulk flow
through possible pinholes in the membrane. Under the membrane are seven l/8
inch diameter stainless steel rods to reduce bowing of the membrane under the
slight hydrostatic pressure.
The membrane transport characteristics were measured under well stirred
conditions in a commercial spiral continuous flow dialysis cell. The values deter-
mined were L = I.4 x IO-’ cm,$/atm and w~,~, = 3.3 x IO-” cm/s. The mem-
brane thickness, measured with a micrometer, was found to be approximately
fifty microns. The cellulose acetate membrane \vas supplied through the courtesy
of ROGA division of Universal Oil Company.
304 J. 0. KESSLER AND C. D. MOODY

The ten molal driving feed solution consisted of five moles of glucose and
fructose in one kilogram of water. Due to the high osmotic pressure of the solution,
it is not subject to microbial attack while in storage. However. four _erams of
benzoic acid were added in order to prekent microbial attack on the membrane,
and twenty-one grtims ofdtric acid and six grams of sodium hydroxide were added
to lower the pH to 4.1.
The refractive index of the driving solution product was measured using a
temperature compensated hand refractometer (American Optical 10430). c,,(h)
was measured with a conductivity bridge. c~(/I) was calculated based on the
measuredc&(h) and the refractometer reading. QJO) was monitored \vith a no. 2

TABLE III

LI(PERIUEhTAL RESULTS OF A LABORATORY 4ORWARD OSMOSIS EVIRAC-TOR

-- --_--- ----.-. .-.- -. ---

/h/& (0) QdO!:lh Qdh?JIh QA (hi .‘Ih Cdll)/C. rdh) Fraction

cm%lg g:cm =s c&s ‘i lOma CnliS :T lo-’ ni0lr.r liter of nurrient
2: IO-4 .< 10-a H-0 lost
---- _---- ---.- -- .__--

I.91 5.24 2.S1 I .99 0.217 1.67 0.20

3.71 270 1.81 1.54 0.262 I .25 0.16
7.07 1.41 1.2 I .06 0.313 0.965 0.18
18.7 0.535 0.707 0.654 0.445 0.701 0.07

L = 1.4 x 1O-5 cm/s atm ih = 436 cm2

)(‘I = 3.3 x 10-s cm/s ~6 = 0.5 cm: )i .= 0.4 cm
cd(O) = IO molal glucose-fructose Awmge sources solution = I on!s
G = 0.6 molal sodium chloride Maximum driving solution velocity -i 0.024 cm’s

IO IO
a EXPERIWENTAL C,,W/C, d EXPERIUEWAL C..(rl/C,
0 EX~ESIUEHTAL IO’rlJh) 0 EXPERIMENTAL IC”s i,lnl
06 - 08 - P.QEDICTED C,,WK, end &ih)
t PREDICTED C&WC. and ,b, 1
y BY MODEL 2 !? 6Y MODEL 3 F(39 4 = 0 oc53cm
2 06 $j 06 t S,.00029cn

~.D;:--r,
4X10' 0 4x104 8x10’ ,2x10’ 16x10’ 20x10’
0 8x10’ 12x10’ 16rl0~20rld

lh/O,to) in
cmhc)
(-
gram
_ fh/C!Jo) m (-)prom -
Fig. 3 and 4. Comparison of experimental and prcdictcd rcsuhs of an experimental
forward osmosis extractor. cdXh),‘c. is tht normalized salt cont. in the nutrient solution. q&h) is
the awzrage membrane water flux in cm/s. Model 2 1.1) nq&cts concentratton polarization. IIY
Fig. 4 6. and 6d are the estimated thicknesses of the concentration boundary layers. See Table III
for the operating parameters.
 DRINKING WATER FROM SEA WATER BY TORWARD OSMOSIS 305

Gilmont flowmetcr. Experimental results arc listed in Table III. The fraction of
glucose and fructose lost is calculated by the equation, 1 - q = 1 - [c,(h)
Qct(WdO) Q.&W-
The experimental results for 3,(/t) and c,,(/r),/c, are plotted wxsw r/!/Q,&O)
in Figs. 3 and 4_

ESPERIMEXI-AL ERROR

The seven membrane support rods which run the length of the extractor arc
an unfortunate source of error in the design of the experimental extractor. The
rods occlude sections of the membrane and also create an unstirreci region in their
vicinity H here one may expect an atypically high concentration of sodium chloride
at the membrane surface. This local increase in sodium chloride concentration not
only decreases the membrane water flux but greatly increases the membrane salt
flux in the area along the rods.

COZlPARlSON OF PREDICTED ASD ESPERIMESTAL RESULTS

In Fi_e. 3 are the results for the experimental extractor as predicted by Model
2 of (/). The average membrane water flux, (5,(/~). is predicted reasonably well
exept at high flo\\ rates. Because concentration polarization is neglected, the
model overestimates the membrane water flux at high flow rates and under-
estimates the membrane salt flus. Both of these factors contribute to an undcr-
estimation of c,,,(h),‘c,, the salt concentration in the nutrient solution.
Model 3. discussed in (I). attempts to Include the fluid dynamics of the
driving aad source solutions by assuming the existence at the membrane of two
unstirred films or concentration boundary layers of thicknesses 3, and 6,. The
concentrations are assumed to equal the bulk solution concentrations, cd(x) and
c,, at the distances ~5~and 6, from the membrane. The concentrationsnear the
membrane surfaces are given by: c,,,(x) = cd(s) exp (- (1,,,(.~)~5~,/i)d(_~)) for the
driving side and CJS) = c, txp (q,(s) b,lD,) for the source side where Dd* and
D, are the diffusivities (cm’/s) of the driving and source solutes. The concentration
boundary layer thickness &, and 6, can be expected to chan_ee with fluid velocities,
viscosity and turbulence. Nevertheless, as a first approximation, average values of
6, and 6, were assumed for the Mode1 3 calculation.
Based on the eight experimental values of cdr(h)/cS and (j,(ll), 6, = 0.0029 cm
and 6, = 0.0453 cm were estimated by nonlinear least squares analysis with the
Marquardt algorithm (9) contained in the FORTRAN model building package

--_-
+ ,4ssuming Dd is inversely proportional ?o ,, (8). !he ditTusivity of glucose and frucrosc in water
may be approximated by the rcla:ion: DJ (cm$k) := 0.673 x IO-’ exp (--.&,I). bd is the
diffusivity of the driving solute avenged in the direction normal to the mcmbranc.
306 J. 0. KESSLER AND C. D. MOODY

XTRACTR (IO). The residual sum of squares (normalized by the respective means
of the experimental values. c&)!c, and 2jm(/r) is 3.4 x 10e3. The model explains
99.674 (corresponding to rz) of the variance of the experimental values_ The
predicted results are shown in Fig. 4.

C0NCLUS10N

Both theoretical and experimental results demonstrate that a forward

osmosis extractor can transfer flesh water from sea water into a solution containing
low molecular weight nutrients. The five liters of water obtained per kilogram of
nutrient, the transfer rate of water through the membrane of > 5 gpd/ft*. the low
loss of nutrients and the simplicity inherent in the forward osmosis process indicate
that fonvard osmosis can be used to obtain emergency water for humans at sea.
The correspondence of theory and experiment is good.

This work was aided by a University of Arizona Institutional Grant and it

Grant from the University of Arizona Foundation. We should also like to thank
Drs. A. R. Kassander and D. B. Thorud for their continuing support and encour-
agement.

REFERENCES

1. C. D. Mooor AND 1. 0. K~LER. Fomwd Osmosis Extractors, De.ditturion. 18(1976) 283.

2. R. E. KRAVA~ AND J. H. DAVIS, Desalination of Sertwatcr by Direct Osmosis. Drsalittariort.
16 (1975) 151.
3. A. V. WOW. Thirst: Physiology of the Ltrge IO Drink and Pro!&ms of IC’ufrr Luck. Charles
C. Thorn- Springfield. 111. 1958, p. 318.
4. S. S~URIRUAS. Reverse Osmosis. Academic Press, New York N.Y.. 1970, p. 229.
5. H. K. LO~DALE. et al., Study of Rejec!ion of Various Solutes by Rcversc Osmosis Mcn-
brana. afire of Saline Warcr, Rcs. Develop. Progr. Rept. h’o. 447, 1969, Table 3.
6. J.‘J. L-~cxs ASD J. V. WAIT, Dart P - 3 Portable CSSL-type Simulation Language. Sittttda-
tion, 24 (1975) 17.
7. U. Mans, Desalittafion &v Rew.-se Osmosis. M.I.T., Cambridge. Mass.. (1966) 100.
8. C. J. GEANKOPL~S.bfu.s.s Truttsporr Phettomctta, Holt, Rinehart and Winston, San Francisco,
C&f., 1972. p. 127.
9. D. L. MARQUARDT, An Algorithm for Least Squnrt.!r Estimation of Non-Linear Para-
mctm, J. Ser. Inl/rrsr. Appl. Math., 2 (1963) 431.
10. J. W. WHY, XTRACTR-Parameter Estimation through Non-Linear Regression, tnternal
Report, College of Mines. University of Arizona.