VOL. 12, NO.

3 WATER RESOURCES RESEARCH JUNE 1976

A New Model for Predictingthe Hydraulic Conductivity

of Unsaturated Porous Media

YECHEZKEL MUALEM

Facultyof Civil Engineering,Technion-lsraelInstituteof Technology,Haifa, Israel

A simpleanalyticmodelis proposedwhichpredictsthe unsaturated

hydraulicconductivity
curvesby
usingthe moisturecontent-capillaryhead curveand the measuredvalueof the hydraulicconductivityat
saturation.It is similar to the Childs and Coilis-George(1950) model but usesa modifiedassumption
concerning
thehydraulic
conductivity
oftheporesequence
in orderto takeinto'account
theeffect
ofthe
larger pore' section.A computational method is derived for the determination of the residual water
content and for the extrapolation of the water content-capillary head curve as measuredin a limited
range.The proposedmodelis comparedwith the existingpracticalmodelsof Averjanov(1950), Wyllie
and Gardner (1958), and Millington and Quirk (1961) on the basisof the measureddata of 45 soils. It
seemsthat the new model is in better agreementwith observations.

INTRODUCTION 0, , and l, respectively.Jacksonet al. [1965], Kunze et al.

The various models used for predictingthe hydraulic con-
ductivity of unsaturated soils were reviewed by Brutsaert
[1967]. We may distinguishbetween two main groups. The
first is based on a generalizationof Kozeny's approach for
saturatedand unsaturatedporousmedia, accordingto which
the relative hydraulic conductivityKr is a power function of
the effective saturation Se, i.e.,
Kr = K/K.,t = Se"
where
[1968], Greenand Corey [1971], and Bruce[1972]havechecked
the reliability of (4). It seemsthat the MQ formula is in
somewhatbetter agreementwith measureddata than the other
formulae.
The three main modelsrepresentedby (1), (3), and (4) do
not seemto have been testedtogether againstmeasureddata
before.
The purposeof the presentstudy is to proposea new sim-
(1) plified model which minimizes the deviations between pre-
dicted and measuredK(O) curves.

& = (o - - (2) THEORY

where O and Or are the actual and the residual water content, We considera homogeneousporousmedium, having inter-
respectively.Following this approach,A verjanov[1950] pro- connectedpores definedby their radius r. The contribution of
posedthe value a = 3.5, whereasIrrnay [1954] derived(1) full poresof radii r - r + dr to 0 is
theoreticallywith a = 3.0. It seemsthat for a wide variety of f(r) dr = dO (5)
soils, a = 3.5 leads to a better agreementwith observations
[Brooksand Corey, 1964;Boreli and Vachaud,1966]. wheref(r) is the porewater distributionfunction.We have
The secondgroup includesthe models of Burdine [1953],
Wyllie and Gardner [1958] (WG in this paper), Farrell and
Larson [1972], and Childsand Coilis-George[1950] (CCG in fnl(r)
raindr=O(R) (6)
thispaper) and the modifications to the CCG modelproposed and in particular,
by Marshall [1958], Millington and Quirk [1961] (MQ in this
paper), and Kunzeet al. [ 1968].The modelsof thisgroup make
useof the measuredcapillary head-water content(O) curve f,m.x
l(r)
rain dr=0,, (7)
to derive the hydraulic conductivityin the unsaturatedstate.
The areal porosityis equal here to the volumetricporosity,so
While in petroleumengineeringthe 'Burdineequation'
f(r) dr representsthe ratio betweenthe poreareaof radii r - r
+ dr and the total area.Considera porousslabof thicknesshx
K,(O
) $ 0
dO/P
0
dO/p
(3) (x -, x 4- hx along the axis). The pore area distributionat the
two slabsidesis identical.For hx >> Rmax,completerandom-
' 0=O--O nessof the relative positionsof the two slab facesis assumed.
The probabilityof poresof radii r - r 4- dr at x encountering
is commonlyused,soil scientistsrefer generallyto a modified
pores of radii p '- p + do at x + Ax is ,,
form of the CCG equation,
K,(O) a(r, p) = f(r)f(p ) dr dp (8)
Here no direct connectionbetweenthe poresr and p doesexist
= S,t
[2.(/--
i)+ 1]/ [2(m
--i)+ 11 along the x axis. The other extreme caseoccurswhen hx - 0.
Then the correlation betweenthe two slab facesis complete.
Here rn representsthe total number of intervals into which the Sincewe are concernedwith the effectof pore changeson the
0 domain is divided, and l is the number of intervals up to a hydraulicconductivity,it is more relevantto considerhx to be
prescribedvalue of 0. CCG, MQ, and Kunze et al. suggest = of the same order of magnitude as the pore radii. Then the
probabilityof the connectionof a porer - r + dr to a porep -,
Copyright 1976by the AmericanGeophysicalUnion. p 4- dp is
513
514 MUALEM:UNSATURATED
FLOW MODELING

a(r, p) = G(R, r, p)f(r)f(p) dr dp (9)

K(O) = Sf dO/q, dO/ (14)
G(R,r, O) is a correction
accounting
for partialcorelation
betweenthe pores r and 0 at a given water content O(R). where n may be positive or negative. Equation (14) is very
The contributionof the actualflow configurationin the slab simple and easy to apply. For (0) given in analytical form,
to the hydraulic conductivitycannot be accuratelyassessed. Kr(O) can be derived explicitly. For example, introducing in
We use,therefore,two simplifyingassumptions: (1) thereis no (14) the expressionused by Brooksand Corey [1964],
bypassflow betweenthe slabpores,and(2) the poreconfigura-
tion may be replacedby a pair of capillaryelements(Figure 1) Se -- (l/cr) -x (15)
whoselengthsare proportional to their radii:
yields
l/l = r/o (1O)
gr(s,)-- S,"+'+vx (16)
The hydraulic conductivity is then found to vary as rp (see
gr(l)-- (llcr) -2-x(2+n) (17)
Appendix 1). If we use a correction factor T(R, r, 0) < 1 to
accountfor eccentricityof the flow path (tortuosityfactor), the With n = 0, (16) and (17) reduceto the equationsobtainedby
contribution of the r - 0 elementto the relative conductivity Brutsaert [1967], using the CCG model. However, the in-
becomes clusionof n (which accountsfor the correlation betweenpores
and for the flow path tortuosity) in (16) and (17) contributesto
d Kr(r, O) a more flexible formula of Kr(Se) and therefore to a greater
chance of agreement between theoretical and experimental
T(R, r, o)G(R, r, o)ro/(r)/(O) dr do curves. Another example is Farrell and Larson's [1972] for-
mula

f;rnax
fmax
T(R
....r,p)G(R
rain ..../',p)rp/(r)/(p)
rain dr
do 1 = lerea('-se) (18)
(11)
which when it is substitutedinto (14) leadsto
For a given 0(R) the correspondingKr(0) is
Kr(S) = S"(e""se- 2e"s + 1)/(e""- 2e" + 1) (19)

Kr(O) =
ruefnT(R,
rain r,p)G(R,
r,p)rp/(r)/(p)
drdomeasured
rain
It is not clearwhether(16) or (19) is in betteragreementwith
data. Equation(16), however,is easierto usein the

f;m..
mifm.
T(R,
r,p)G(R,
r,p)rp/(r)/(O)
drdo
rain
derivationof analyticalsolutionsof complicatedunsaturated
flows.
(12)
COMPUTED RESULTS
Sincethere existsno procedurefor an independentdetermi-
Beforeagreementof theory with measurements
and com-
nation of T(R, r, O) and G(R, r, 0), we assumewith Burdine
parisonwith other methodscan be examined,the residual
[1953]and M Q that the tortuosityand correlationfactorsare
water content{Drand the power n of Se ((14)) must be deter-
power functionsof 0, thus dependingsolely on R. Hence
mined.The influenceof usingpartial and completeb-Ocurves
on thecomputedhydraulicconductivitywascheckedby Kunze
r/(r) dr Of(p) dp et al. [1968].It seemsthat completeb-Oinformationimproves
K(O) = Sf rnin rain the quality of the predictionmainly as a resultof a better

fRmax
f/emax
Rrnin
rf(r) dr
rain
p/(p) dp fulfillmentof the requirementthat K = 0 for O = {Dr.They
recommend,therefore,extrapolationof the measuredportion
of the b-O curve. During the presentstudy we realizedthat

1
r/(r) dr becauseof the strongsensitivityof the computedK(O) curve
to the value of Or, the decision about which model compares
= S," i__..
_ (13)
more favorablywith experimentaldata of a givensoil is, in
fact, governedby {Dr. For this reason,we think that it is
necessaryto usea standardanalyticalprocedureto fix a value
Applying the capillary law r = C/ and (5) to (13), one of {Dr and to extrapolate a partially given b-{Dcurve. We
obtains
proposehereina convenientprocedure,basedon the assump-
tion that the soil characteristiccurvein the extrapolationrange
can be analyticallyrepresented by (15), whichfulfilsthe con-
dition b- Dofor O {Dr.Parameters{Drand X are computed
with the aid of a minimum squaredeviationprocedurefor
regression of measuredb-Opointsto (15) in the rangeO < Op,
2r 2P whereOp is the value at whichthe measuredcurveshowsan
inflectionpoint (Appendix 2).
Now if we perform a similar fitting procedurefor the Kr(O)
curveto (14), we can expectto get differentvaluesof n for
differentsoils.As was alreadymentioned,the value of n may
be negativetoo. Of course,thisposesa greatdifficultyif one is
interestedin a universalKr(O)representation,valid for all soils.
Fig. 1. A combination of cylindrical tubes used to evaluate the As a sort of compromisebetweenboth representations, it is
hydraulicconductivityof a pair of capillaryelemehts. suggested that an expressionbe adoptedfor whichthe square
 MUALEM:
UNSATURATED
FLOWMODELING 515

deviation,
averaged
overa greatnumberof soils,isminimum. 0.5 may indeedhold as the best value. Hence the suggested
In essence,n becomesan experimentallydetermined parame- formula for Kr(S) becomes
ter.

Forty-five soils for which measured-O and K-O (or K-C)

data on drainage are available in the literature were usedfor
computation. In Table 1 a list of the soils, their measured
(mln,Pmln,and Omax,the computedOr obtainedby usingthe
proceduresuggestedin Appendix 2, and the power 3, of Se are
given. For each soil the mean squaredeviationD betweenthe
measuredhydraulicconductivitiesKrmand the computedones
Krc,
I ,.Re I /2
In practice,we distinguishbetweenthe extrapolatedregion0
_< S ( S mm( ( (Omm), for which an analytical
expression ((15)) is usedto representthe f-S curve,and the
measuredregionS mm S 1, wherethe computationis
carriedout numerically.Continuity requiresthat the extrapo-
lated curve meet the measuredone at (from,S ram).Thus (21)
becomes

D-- e Inin [ln(K,c)

-- In(K,m)]"
(1-- min ) (20)
is computed.Eight differentvaluesof n wereused(substituted
in (14)): n - -1 + 0.5j,j = 0, 1, ". , 7. In orderto maintain
consistencyin the numericalcomputationof D for all soilswe self2 Ie in__-- (22)
have used a constant increment ASe = 0.02. In Table 2 the
meanvalue/, the standarddeviationa, and the coefficientof Semin/(1
1/)mi
n fS dSe/ e rain

variance e for the 45 soils are given as a function of n. The

continuous
graphof/(n) asplottedin Figure2 shows
thatn = while the WG model ((3)) yields

TABLE 1. Computed Or and X for the 45 Soils

Soil /min
No. Index References {max {min cm HO

1 lOO6 BeitNetofaclay[Rawitz,1965] 0.446 0.241 1.51 X 104 0.010 0.19

2 1101 Shluhotsilty Clay[Rawitz, 1965] 0.385 0.163 1.51 X 104 0.010 0.20
3 2002 Silt Mont Cenis[Vachaud,1966] 0.447 0.042 1.70 X 105 0.010 0.36
4 2004 Slatedust[ChildsandCoilis-George, 1950] 0.482 0.110 1.44 X 102 0.090 5.69
5 3001 Well siltyclayloam [JensenandHanks, 1967] 0.470 0.140 2.12 X 102 0.090 1.52
6 3002 Amarillo silty clay loam [Brooksand Corey, 1966] 0.455 0.140 2.25 X 102 0.110 2.35
7 3101 Rideauclayloam[Topp,1971] 0.416 0.286 4.19 X 102 0.280 1.62
8 3301 Caribousilt loam [Topp,1971] 0.441 0.313 4.25 X 102 0.280 0.91
9 3302 Grenville silt loam [Staple,1965] 0.475 0.037 1.00 X 10 0.010 0.34
lO 3304 Touchetsilt loam [Jensen andHanks, !967] 0.480 0.170 2.35 X 102 0.120 1.71
11 3305 Ida silt loam(>15 cm) [Greenet al., 1964] 0.530 0.175 2.00 X 10a 0.060 0.38
12 3306 Ida siltloam(0-15cm)[Green
et al., 1964] 0.554 0.219 2.00 X 10a 0.010 0.27
13
3307 Touchet
siltloam(GeneralElectric3) [Brooks
andCorey1964] 0.469 0.180 4.14 X 102 0.130 1.89
14 3403 Pachappaloam[Jackson
et al., 1965] 0.456 0.007 3.19 X 10 0.002 0.42
15 3404 Adelanto
loam[Jackson
etal., 1965] 0.426 0.012 4.65 X 106 0.007 0.50
16 3405 Indio loam [Gardner,1959] 0.450 0.021 1.50 X 106 0.010 0.81
17 3407 Guelphloam[ElrickandBowman, 1964] 0.520 0.236 1.00 X 10a 0.130 0.41
18 3501 Rubiconsandylam[Topp,1969] 0.381 0.166 2.40 X 102 0.150 2.08
19 3503 Pachappafinesandyclay[ElrickandBowman,
1964] 0.334 0.049 1.50 X 10' 0.030 0.44
20 3504 Gilat sandyloam[Hadas,1967] 0.440 0.130 1.02 X 10a 0.010 0.44
21 4106 Sand [PoulOVassilis,
1970] 0.272 0.090 3.60 X 10 0.010 1.83
22 4107 Sand[Poulovassili
s, 1970] 0.258 0.084 3.80 X 10 0.010 2.87
23 4109 Botany
sand
fraction
(150-300
m) iWatson,
1967] 0.350 0.055 5.70 X 10 0.040 8.35
24 4111 River sand(screened)[Jensen
andHanks, 1967] 0.400 0.060 1.50 X 102 0.050 1.57
25 4114 Volcanic
sand[Jensen
andHanks,1967] 0.350 0.050 1.85 X 102 0.040 1.30
26 4116 Sandfraction(150-300/m)[Kastelanek,
1971] 0.372 0.045 8.00 X 10 0.040 4.94
27 4118 Sablede rivire [Vachaud,
1966] 0.342 0.075 1.90 X 102 0.060 0.92
28 4120 Gilatrinesand [Rawitz,1965] 0.179 0.070 1.51 X 10 0.010 0.27
29 4121 Rehovotsand[Hadas,1967] 0.400 0.020 2.50 X 10a
.
0,015 0.83
30 4123 P0uderRiversand[BrooksandCorey,1966] 0.364 0.044 8.20 X 10 0.030 2.92
31 4126 MolongloRiversand[Talsma,1970] 0.277 0.098 3.00 X 10 0.010 0.96
32 4129 BeitDagansand[Rawitz,
1965] 0.161 0.052 1.49 X 10 0.040 0.37
33 4130 Hygienesandstone[BrooksandCorey,1964] 0.250 0.151 2.01 X 10 ' 0.140. 3.78
34 4131 Bereasandstone
[Brooks
andCorey,1964] 0.206 0.064 2.34 X 10' 0.050 2.13
35 4132 Fragmented
FoxHill sandstone
[Brooks
andCorey,1964] 0.503 0.166 1.16 X 102 0.160 2.61
36 4133 Fine sand(General Electric 13) [Brooksand Corey, 1964] 0.356 0.063 3.02 X .102 0.050 1.98
37 4134 Volcanicsand[Brooks
andCorey,1964] 0.365 0.058 2.73 107 0.050 1.65
38 4137 Sandfraction(150-300tm) [Watson,1967] 0.350 0.056 5.70 10 0.050 11.67
39 4141 Sandfraction(1.0-0.5mm) [ChildsandCoilis-George,
1950] 0.357 0.034 3.64 10 0.020 2.80
40 4142 Sandfraction(0.5-0,25mm) [ChildsandCollis-George,
1950] 0.364 0.040 4.40 X 10 0.030 5.69
41 4143 Fragmentedmixture[BrooksandCorey,1964] 0.437 0.134 1.07 10 ' 0.120 2.65
42 4147 Plainfieldsand(25-60 cm) [Blacket al., 1969] 0.307 0.060 2.05 102 0.050 1.45
43 5002 Glassbeads
[Brooks
andCorey,1964] 0.383 0.037 3.01 102 0,030 1.90
44 5003 Aggregatedglassbeads[ToppandMiller, 1966] 0.548 0.080 8.26 10 0.060 3.57
45 5004 Monodispersed glassbeads[Toppand Miller, 1966] 0.326 0.033 6.82 X 10 0.020 6.24

O.... Omin,and 19rare givenin percentof total volume.

 516 MUALEM:
UNSATURATED
FLOWMODELING

TABLE2. Computed
D, a, ande for EightValues
of n Obtained
by Using(14)

- 1.0 - 0.5 0.0 0.5 1.0 1.5 2.0 2.5

15= ']Dd45 1.40 1.18 1.01 0.97 1.09 1.32 1.62 1.94
a = [(Dr-/5)2/45]/2 0.73 0.70 0.75 0.82 0.89 0.96 1.04 1.58
= a/I5 0.52 0.60 0.74 0.84 0.81 0.72 0.64 1.13

effective
watercontentand the effective
saturation,
respec-
tively.

= s.
Semin/(1+ 2/X)P,,,i,-I- 2 fS
$e
' "
dS./ 2 Thenumerical
(23) assumption
procedure
thatthemeasured
usedin thisstudyisbasedon the
p-Se
curvecanbeveryclosely
S....i/(l + 2/X),,,, + dS./ approximated
by a continuous
polygon,i.e., that pcanbe
e rain expressedby
Instead of presentingthe MQ formula based on the CCG
modelin the form of finite sums,as is usuallydone in the = ,- (S,,+-
(,- S,,)(S,- S,,) (26)
literature,we believethat it isworthwhileto presenttheK(S,)
relationshipin termsof integralsfor severalreasons:(1) we S,i _< S, < S,i+
cancomparedifferentformulaein an easierway,(2) if analyti-
cal relationshipsbetween and S, are available, the K-S, In thiscasewearenotlimitedto usingconstant intervals,
the
relationship
mayalsobeobtainedin a closedform,and(3) the node points areconveniently
chosen withregard to thecurva-
computerpermitsuseof a varietyof procedures for replacing ture of the givengraph,and the computation of Kr(S,) is
integralsby finite sums.This latter argumentis of extreme accurately performed.When(26) is used,(22) becomes
importance,as will be shownlater. For this reason,we have
elaboratedthe CCG model(seeAppendix3) to obtaina final
compactintegralform of the K-O relationship:
Kr(S,)
=S,'/2{[S,
min/(l
+l/X)min
K(O) = &"
, ,+/ !n
fo(0 -- O)dO/l?/ fo
"st
(O,,,t-- O)dO/l," (24)

Usinga = 4/3 and (15) to express

the-Se dependence
on the
extrapolated portion, we have

Kr(Se)__
Se4/3
I l (._Se.W.e
.'e.mi.n__
2
min2+2/x/
x+5) S whiletheWG andthemodified
MQ model((23)and(2))
yield

+ (s, - s) ds/
e i !1

K(S,)
=S,'{[S,
min/(l
+2/X)min'
T.Z '5'?x- 2 + 2/x/
, Ji+1 So
rain/(1
+ 2/X)min
2
+ ( - s) ds/ (25)
e m i n

whichis theequivalentformulafor the MQ model.Parameters

+1Sei+l
--Sell}
--I .... (28)

0 and s (in (24) and (25)) are demivariablesrepresentingthe

and

2.0

1.8 Kr(S,) = S,4/3 Serain

2
1 + 2/x 2+
1.6

+
I Si+
Si S,

+ ('// ///+
si+! '-_ ..s.,_. lB
1.0
- %+....
+ I +

0.6 I
-I.0 - 03
I
0.0
I I
03
I
1.0
I I
1.5 2.0 2.5
I
Se
min
2 I si (s,_.+
--si)
1 1

Fig. 2. Computed/5
based
on45 soilsasa function
of thepowern
(see (14)). ,.)+(,,.+:-_
::). (29)
 MUALEM'. UNSATURATEDFLOW MODELING 517

respectively.In the following the suggestedmodel ((27)), as The presentation of the CCG model is modified to derive a
well as the modelsof Averjanov, WG, and MQ ((1), (28), and compact integral formula instead of its usual form in finite
(29), respectively),will be comparedwith the experimentally sum. This improvement enablesus to use the same modified
measured data. proceduresin applying the various modelsand therefore to
In Table 3 the deviation D between the measuredKr(Se) perform a more reliablecomparisonamongthe models.
curves and the computed ones (using the proposed method For computational benefit the measuredb-O curve is re-
and the three methodsmentionedabove) is given for each one gardedas a continuouspolygonconnectingthe given (b,O)
of the 45 soils. The italicized values indicate the model which points. This approximationenstresan accuratecomputation
yields the minimum deviation betweenthe measuredcurves of Kr(Se)withouttruncationerrors(whicharequitesignificant
andthe forecasted ones.The averagevalueD for all 45 soils, for valuesof b near zero).
the standard deviation a, and the coefficient of variance e are The proposedmodel, as well as the three main existing
given in the last three linesof the table. It may be seenthat for modelsof Averjanov, MQ, and WG, is comparedwith mea-
various soils, best resultsare achievedby different methods. sureddataof varioussoils.Sincethistestis carriedout against
This can also be concludedfrom Figures3a-3p, in which the data the accuracyof which is undetermined,45 soilsof differ-
computedresultsare shown for a sample of 16 soils. On the ent types are considered to ensure solid conclusions. It is
other hand, comparison between the different models on an
TABLE 3. Deyiation D Betweenthe Measuredand Computed
overall basisshowsthat best resultsare achievedby the pro- Kr(Se) Obtainedby Using the Models of Averjanov, WG, and MQ
posedmodel.This is reflectedby the numberof soilsfor which and the Proposed Model
D is minimum(seeTable 3 and Figures3f, 3g, 3j, and 3m-3p)
and by the computedaveragevalueD for the45 soils.
Very often the agreement of Averjanov's model with the Soil Proposed
No. Index Averjanov WG MQ Model
experimentaldata is quite poor (as it is in soils 1006, 3302,
3305, 3306, 3403, 3404, 3405, and 3407; seealso Figures3a and I 1006 1.94 0.20 0.24 0.29
3c-3f), while in somecases(soils2002,4121, and 4126;see 2 1101 1.26 0.74 0.76 0.64
Figures3b and 3/) a definiteimprovementin resultsis achieved 3 2002 1.72 6.0 l 4.38 4.17
by this method as compared with the other three. It seems, 4 2004 0.22 0.14 0.19 0.40
5 3001 1.08 0.37 0.23 0.76
therefore,that the.b-0curveincludessomeinherentcharacter-
6 3002 0.51 0.36 0.25 0146
istic properties of the soil, which the generalized K ozeny- 7 3101 0.59 !.01 1.18 0.58
Averjanov-lrmay approachignores.One may conclude,how- 8 3301 0.59 2.44 1.48 1.22
ever, that Averjanov's model fits sandswell, while it fails to 9 3302 5.67 1.07 1.69 1.72
10 3304 0.63 0.49 0.58 0,42
describethe Kr-O relationshipaccuratelyfor heaviersoils.Yet
11 3305- 5.12 1.13 1.28 1.34
the Averjanov-lrmay type of equation,with adjustablepower, 12 3306 6.11 2.36 1.26 1.34
is an expressionthat has beenprovedto be very convenientin 13 3307 0.46 0.33 0.44 0.38
the analytical solution of partly saturatedflow problems.It 14 3403 4.38 1.38 1.15 0.94
might be possibleto improve predictionof K(O) by adjusting 15 3404 4.87 2.82 1.59 1.33
16 3405 5.46 1.00 2.23 2.18
the powerfor tortuosityand for X (aswasdoneby Brooksand 17 3407 1.96 0.82 0.44 0.25
Corey [1964] or as shown here in (16)). 18 3501 0.42 0.86 1.04 0.40
Among the other three models the WG model yields the 19 3503 2.18 4.33 2.01 2.25
poorestresults.For soilsin whichdO/db 0 as 0, thereis 20 3504 2.66 2.93 1.61 1.67
21 4106 0.50 1.00 1.03 0.65
an abrupt fall of the computed hydraulic conductivity near
22 4107 0.38 0.62 0.71 0.30
saturation (Figures 3b, 3d, 3e, 3h-31, and 3n). For some soils 23 4109 1.12 0.88 1.18 0.56
thesecomputedresultsare justified by the experimentaldata 24 4111 1.13 2.03 1.82 1.01
(Figure 3e), but in most casesthe contrary is sustained,since 25 4114 1.18 0.92 0.59 0.77
the observeddrop is milder. 26 4 i 16 0.55 0.54 0.60 0.97
27 4118 0.52 1.19 1.73 0.86
Finally, we believethat there is a good chanceof improving
28 4120 0.66 0.80 1.02 0.79
the prediction of K(O) by developingprocedureswhich cir- 29 4121 0.27 3.53 2.77 2.41
cumvent some of the existing limitations. One might neglect 30 4123 0.80 0.70 0.53 0.97
the measured data of -O near saturation and fix a Clear air 31 4126 0.28 2.15 1.21 1.22
32 4129 0.88 2.65 2.65 2.26
entry value. Another modificationis the derivationand useof
33 4130 0.14 0.24 0.26 0.27
an experimental correlation between the value of n in the 34 4131 0.56 0.43 0.62
0.34
power function ((1)) and some physical parameters of the 35 4132 2.85 3.53 3.66 2.91
soils. 36 4133 0.37 0,41 0.36 0.44
SUMMARY AND CONCLUSIONS 37 4134 0.31 0.35 0.56 0.30
38 4137 0,53 0.28 0.53 016
The new model for prediction of K(), proposedherein, is
39 4141 1,38 1.60 1.71 0.93
basedon a reasonableapproximateevaluationof the hydraulic 40 4142 0.65 0,99 0.62 0.91
conductivityof a pore domain with varying shape.The KA) 41 4143 0.52 0.73 0.92 0.41
expressionis derived in a simple integral form. Thus in cases 42 4147 0,71 0.50 0.70 0.37
43 5002 0.91 0.98 1.12 0.43
where the - dependenceis given by analytical formulae,
44 5003 0.61 0.69 1.04 0.21
K(O) can be reducedto a closedform. Very rarely is the 45 5004 0.90 0.80 1.04 0.30
curvemeasuredin the whole range.Hence in order to make /J 1.49 1.32 1.17 0.97
the various models more practical tools an analytical pro- a 1.64 1.22 0.88 0.82
cedureis suggestedfor the determinationof the residualwater = a/D 1.10 0.93 0.75 0.84

contentand the extrapolation of the - curve into the range The italicizedvaluesindicate the model which yieldsthe minimum
for which no measured data are available. deviation between the measured curves and the forecasted ones.
 0 0
I0 I0 10

-1

-2
I0
Kr Kr
I
3 , -
2

-3
I0
0.30 0.3. 0.38 0..2
(vournetric)

Fig, 3a

0'2(0voiflt
r,f/"O _s
I0 0.200.30O.Z.O
0.50
' Fig.
3b (v;i.rn3
r,
c)
r

K I) --

r, //,- 1
G5 3/.05 I0
y 3z.
07--
o. oito,uOer,,o.
o.o ''o.,o
o.oo.,oo.o
Fig.
3g (volumetric) 10
0.2 0.32 0.0 0.8
(volumetric)
Fig.3e Fig.
3/
io
-1
I0
-1
IO

-2
IO

31Ol i)3
-& 3301

-5

0.30 0.34 0.38 O.&2 -5

(volumetric) 10 I I I ,
0.30 0.34 0.38 O.&2 O.&6 0.10 0.20 0.30 O.&O
Fig.3g (votumetric) ( votumetric)
Fig. 3h
Fig. 3i

Fig.3. Computed
relative
hydraulic conductivities
for16soilsusing
fourmodels:
Averjanov's
(Y symbols;
see
(1)),
WG's(plusses;
see(28)),MQ's(solid
triangles;
see(29)),andtheproposed
model
(solid
circles;
see(27)).Themeasured
curveis shownby solidsquares.
iO i0

-2 -2
I0 I0

-3
-3 -3
I0 I0

&Ill all/.,

-&
-
IO I I I i I I
-&
I0 0.10 0.1/-. 0.18 0.22 0.26
(votumetric)

-5 Fig. 31
-5 I0
I0
0.12 0.20 0.28 0.36 0.0. 0.12 0.20 0.28 0.36
(votumetric) (votumetric)

Fig. 3j Fig. 3k

00 -- o

o ! ii, I '

IO0.o&
0.12 0.20 0.28 0.36
(votumetric)

Fig.3m !-5
0.0/-, 0.12 0.20 0.2B 0.36
o.,o
Oi,vOo,uO2,,o
o.o Fig. 30
(votumetric)
_

Fig. 3n

-5
I0
O.O& 0.12 0.20 0.28 0.36
(votumetric)

Fig. 3

Fig. 3. (continued)
520 MUALEM:UNSATURATED
FLOWMODELING

shown that the proposed model is more reliable than the In this work, (15) is adopted to representthe extrapolated
existingmodelsand clearlyimprovesthe predictionof unsatu- part of the -{D curve. Demandingthat the extrapolatedcurve
rated hydraulic conducti,ities. shouldpassthroughthe measuredlast point (ml,, {Dmin)leads
to

APPENDIX 1' HYDRAULIC CONDUCTIVITY

OF THE CAPILLARY ELEMENT

In mostof the rationalmodelsfor predictingK oneconsid-

Se'rnin
-- Omin_
Or= (AS)
ersthe conductivityof a pair (or a series)of capillaryelements. or

Relyingon Poiseuille's
equation,ChildsandCoilis-George
In (m,./k) = X- In (S,/S, Era) (A9)
[1950] assumedthat the conductivityis determinedby the
radiusof the narrowerelement,whereasWyllie and Gardner On a log scale. (9) describesa straight line. As a matter of
[1958] useda reducedpore radius. Here a different relation- convenience we define
ship is adopted, basedon the pore configurationshownin
Figure 1. y = In (mln/) X = In (S/Smln) (A10)
To simplifymatters,we disregardspecialeffectsand assume Sincethe extrapolated curve should match the measuredone,
that eachcapillaryin Figure 1 obeysthe Poiseuilleequation, we demand that the dispersionof the measuredpoints, up to
i.e., the inflection point (Om < < ), around the analytic
4 curve(A9) shouldbe minimum. The sum of the squaredevia-
rr4gdc rr g tion of the measureddata from the analyticcurve is
Q= 8 dx Q'= 8 dx (A1) N

d= [y, -- y(xi)]
where = + z is the head,, is the liquid kinematicviscosity, 1

and Q is the discharge.For this combinationin series,the total

head loss is
= y, -- y,x,+x, (All)
A = A + Aq%. (A2)
and requiringd to be minimum (ad/aA = 0), one obtains
while the dischargeis
Q = Q = Q: (A3)
X= In S. / , ln(i,ln
x S. (A12)
If the elementof Figure 1 is replacedby an equivalenttube
and
of radiusR and lengthL. thenthe dischargerelationshipyields
]{4 A/L = P Aqbl/l= p4A:/I.
while equivalencyof volumesgives
(A4)
N (min
2 1(min
lB
(
RaL = 11 + pla (A5)
, in (Al)
Assumingmoreoverthat the lengthsare proportionalto the
radii [Fatt, 1956], If we assumea seriesvalue Orj '- 0.01/,j = 1, 2, '. ', up to
l/l = r/p (A6) {Dm,,the corresponding
Xj anddj are obtainedby using(A 12)
and (A13). The residual water content {Dr is chosen as the
we obtain
valueof {Drwhichyieldsthe minimumvalueof d.
R = rp (A7) APPENDIX 3: MODIFIED FORMULATION
OF THE CCG MODEL
This meansthat the 'large pore' has a more important influ-
ence than is generallyassumed.The reasonis, of course,that The basic equation suggestedby Childsand Coilis-George
we have taken into accountits lengthand not only its cross [1950] as set forth by Brutsaert [1967] is
section.

APPENDIX 2: PROCEDURE FOR DETERMINATION OF {Dr

= Mj'=R
K(R) fr=r2](p)](r)
drdp p= R m in =R min

AND THE EXTRAPOLATED

I-{D CURVE
+ M p"f(r)/(p)dp dr (A14)
p R in P
The determination of {Dris a prerequisitefor usingany of m

the methodssuggested for predictingthe K-{Drelationship.We It is obvious that the integration is carried over the square
define{Dras the residualwater content for d{D/&k 0 for {D domain OABC in the (r, p) plane (Figure 4). The first integral
{Drbecauseit fulfils the other basicrequirementthat K({Dr) = (left-handside) of (A14) is carried over the triangleOBC, and
0. Very often, only part of the -{D curve is measured,{Dris an the secondintegral over the complementarytriangle OAB. By
unknownparameter,and it is not clearhow to extrapolatethe a changein the order of integration,
measuredcurve. This problem becomesembarrassingwhen
comparisonsof variousmodelswith measurements are stud-
ied, becausejust by choosing{Drwe may improve or worsen
one method in relation to the others. There is an intense need
,-e r=,r2f(p)f(r)
p= R m drdp
in =R min

for an objectiveanalyticprocedurefor extrapolatingthe mea-

sured k-{Dcurve.
=
frr,=R
fpp=R
r9'/(r)/(p)dp dr
:=R min
(A15)
 MUALEM:
UNSATURATED
FLOW
MODELING 521

T tortuosity factor.
a,/ constant power.
, r actualand residualwater content.
}max,
}mln measured
maximumand minimumvaluesof
O.
0 = - r, effective water content.
t> radius.
b capillaryhead.
kc, capillaryheadat whichdO/&k> O.
mln minimummeasured valueof
Acknowledgment.
Thepresent
workhasbeensupported
by the
/ U.S.-IsraelScienceBinationalFoundationundergrant442.

A REFERENCES

R Averjanov,
S.F.,About
permeability
ofsubsurface
soils
in case
of
incomplete
saturation,
Eng.Collect.,
7, 1950.
Black,
T. A.,W. R. Gardner,
andG. W.Thirtell,
Thepredictionof
Fig.4. Description
of theintegration
domain
of (A14). evaporation,
drainage,
andsoilwater
storage
fora baresoil,Soil
Sci. Soc.Amer. Proc.,33, 655-660, 1969.
Boreli,
M., andG. Vachaud,
Notesurladetermination
delateneur
en
and a changeof variables, eaur6siduelle
et surla variation
dela perm6abilit6
relative
dansles
solsnonsatur6s,C. R. Acad.Sci.,263,698-701,1966.
Brooks,R. H., andA. T. Corey,Hydraulic properties
of porous
f,r==R
fpp==R
r r2/(r)f(p)
w
dpdr
/i m in
media,
Brooks,
Hydrol.Pap.3, Colo.StateUniv.,FortCollins,
R. H.,andA.T. Corey,Properties
ofporous
1964.
media affecting
fluidflow,J. lrrig.Drain.Div.Amer.Soc.CivilEng.,92(IR2),
-- p2/(p)/(r)dr dp (A16) 61-88, 1966.
in
Bruce,
R. R., Hydraulic
conductivity
evaluation
of thesoilprofile
fromsoilwaterretention
relations,
SoilSci.Soc.Amer.Proc.,36,
it follows
thatthetwointegrals
of(A14)areidentical.
Hence 555-560, 1972.
Brutsaert,
W.,Some
methods
ofcalculating
unsaturated
permeability,
K(R) = 2M pef(p)f(r)
drdp Trans. ASAE, 10, 400-404, 1967.
in r==p Burdine,
N. T., Relative
permeability
calculation
fromsizedistribu-
tion data,Trans.AIME, 198,71-78, 1953.
(A17) Childs,
E. C., andN. Coilis-George,Thepermeability of porous
--- 2M
p=R
p==R rain
pe/(p) /(r) dr dp materials,
Elrick,
Proc.Roy.Soc.,Set.A, 201,392-405,1950.
D. E.,andD. H. Bowman,Noteonanimproved apparatus.for
By definition,we have soilmoisture
measurements,
SoilSci.Soc.Amer.Proc.,28,450-453,
1964.
Farrell, D. A., andW. E. Larson,Modeling theporestructure
of
f,Rf(r)
dr=O(R)
/min (A
18) porous media, Water
Fatt,I., Thenetwork
Resour.
model
Res.,8, 699-706,
of porousmedia,
1972.
1,2, 3, Trans.
AIME,
and substituting
in (A17) we obtain 207, 144-181, 1956.
Gardner,
W. R., Mathematics
of isothermal
waterconduction
in un-
0
saturated
soils,
Rep.40,pp.78-87,HighwayRes.Board,
Nat.Res.
Counc.,Washington,D.C., 1959.
Green,
R. E.,andJ.C. Corey,
Calculation
ofhydraulic
conductivity:
A furtherevaluation
of somepredictive
methods,
Soil$ci. Soc.
Bymatching
atsaturation,
theCCGformula
canbewritten
as Amer. Proc., 35, 3-8, 1971.
Green,R. E., R. J. Hanks,
andW. E. Larson,
Estimates
of field
infiltration
bynumericalsolution
ofthemoisture
flowequation,
Soil
Sci. Soc.Amer. Proc.,28, 15-19, 1964.
Hadas,A., Evaporation anddryingprocess in layered soils(in
Theformulae of Millington
andQuirk[1961]andKunze et al. Hebrew), Ph.D.thesis,Hebrew Univ.,Rehovot,
Israel,1967.
[1968]maybewritten similarly
bymultiplying
theright-handIrmay,S.,Onthehydraulic conductivity
of unsaturatedsoils,
Eos
sideof (A19)bySe',withn = andn = 1,respectively. Trans. AGU, 35, 463-467, 1954.
Jackson, R. D.,R.J.Reginato, andC.H. M. VanBavel,
Comparison
NOTATION of measuredandcalculated
hydraulicconductivities
of unsaturated
soils,WaterResour.
Res.,(3), 375-380,1965.
a probability. Jensen,
M. E., andR. J. Hanks,
Nonsteady
statedrainage
from
f(r) porewaterdistribution
function. porous
media,
J.lrrig.Drain.
Div.Amer.
Soc.
Civil
Eng.,
93(IR3),
G correlation factor. 209-231, 1967.
K hydraulicconductivity. Kastelanek,
F., Numerical
simulation
technique
forvertical
drainage,
J. Hydrol.,14, 213-232,1971.
Kr,K,c relativehydraulic
conductivity
andcomputed Kunze,
R. J.,G. Uehara,
andK. Graham,
Factors
important
inthe
value of calculation
ofhydraulic
conductivity,
SoilSci.Soc.Amer.
Proc..,
32,
I index. 760-765, 1968.
L, l length. Marshall,
T. J.,A relation
between
permeability
andsizedistribution
m index. of pores,J. SoilSci.,.9, 1-8, 1958.
Millington,
R. J., andJ.P. Quirk,Permeability
of porous
solids,
n constant power. Trans.FaradaySoc.,57, 1200-1206,1961.
R, r radius. Nielsen,
D. R., D. Kirkham,
andE. R. Perrier,
Soilcapillary
Con-
Se,Semln effective
saturation
andminimum
effective
sat- ductivity:
Comparison ofmeasured
andcalculatedvalues,
SoilSci.
uration. Soc. Amer. Proc., 24, 157-160, 1960.
522 MUALEM: UNSATURATED FLOW MODELING
Poulovassilis,A., Hysteresisof pore water in granularporousbodies,
Soil Sci., 109, 5-12, 1970.
Rawitz, E., The influenceof a number of environmental factors on the
availability of soil moisture to plants (in Hebrew), Ph.D. thesis,
Hebrew Univ., Rehovot, Israel, 1965.
Staple, W. J., Moisture tension, diffusivity and drying, Can. J. Soil
Sci., 45, 78-85, 1965.
Talsma, T., Hysteresisin two sands and the independentdomain
model, Water Resour.Res., 6(3), 964-970, 1970.
Topp, G. C., Soil water hysteresismeasuredin a sandyloam compared
with the hystereticdomain model, Soil Sci. Soc. Amer. Proc., 33,
645-651, 1969.
Topp, G. C., Soil water hysteresisin silt loam and clay loam soils,
Water Resour.Res., 7(4), 914-920, 1971.
Topp, G. C., and E. E. Miller, Hysteresismoisturecharacteristicsand
hydraulicconductivitiesfor glass-beadmedia, Soil Sci. Soc. Amer.
Proc., 30, 156-162, 1966.
Vachaud,G., V6rification de la 1oide Darcy generalis6eet d6termina-
tion de la conductivit6capillairepartir d'uneinfiltrationhorizontal,
Ass. Int. Hydrol. Sci. Wageningen,82, 277-292, 1966.
Watson, K. K., Experimentaland numericalstudy of column drain-
age, J. Hydraul. Div. Amer. Soc. Civil Eng., 93, 1-15, 1967.
Wyllie, M. R. J., and G. H. F. Gardner, The generalizedK ozeney-
Carman equation, World Oil, 146, 210-228, 1958.
(Received February 21, 1975;
revisedAugust 11, 1975;
acceptedAugust 15, 1975.)