Professional Documents
Culture Documents
The soil-water characteristic curve can be used to estimate various parameters used to describe unsaturated soil
behaviour. A general equation for the soil-water characteristic curve is proposed. A nonlinear, least-squares computer
program is used to determine the best-fit parameters for experimental data presented in the literature. The equa-
tion is based on the assumption that the shape of the soil-water characteristic curve is dependent upon the pore-size
Can. Geotech. J. Downloaded from www.nrcresearchpress.com by Depository Services Program on 02/27/13
distribution of the soil (i.e., the desaturation is a function of the pore-size distribution). The equation has the form
of an integrated frequency distribution curve. The equation provides a good fit for sand, silt, and clay soils over the
entire suction range from 0 to 10"Pa.
Key words: soil-water characteristic curve, pore-size distribution, nonlinear curve fitting, soil suction, water
content.
La courbe caractkristique sol-eau peut Stre utilisCe pour estimer divers paramktres dCcrivant le comportement
d'un sol non saturk. On propose ici une Cquation pour cette courbe caractkristique sol-eau. Un programme non
IinCaire, par moindres cam&, est utilisk pour dkterrniner les paramktres qui permettent d'approcher au mieux les donnkes
expCrimentales recueillies dans la IittCrature. L'Cquation est basCe sur l'hypothkse que la forme de la courbe
caractkristique sol-eau dCpend de la rkpartition de la taille des pores du sol (i savoir que la perte de saturation est
une fonction de cette repartition). L'Cquation a la forme d'une intCgrale de courbe de rkpartition, de frCquences.
Cette Cquation permet un bon ajustement pour les sols sableux, silteux et argileux sur toute la gamme des valeurs
de succion, de 0 I? 10"Pa.
Mots clis : courbe caractCristique sol-eau, rkpartition de la taille des pores, ajustement non linkaire, succion dans
le sol, teneur en eau.
[Traduit par la rCdaction]
For personal use only.
Introduction range. This paper proposes a new equation that can be used
A theoretical framework for unsaturated soil mechanics to fit laboratory data over the entire soil suction range.
has been established over the past two decades. The con- A mathematical basis for the equation is described and a
stitutive equations for volume change, shear strength, and best-fit procedure is outlined to obtain the parameters for
flow for unsaturated soil have become generally accepted the equation.
in geotechnical engineering (Fredlund and Rahardjo 1993a).
The measurement of soil parameters for the unsaturated soil
constitutive models, however, remains a demanding labo- The soil-water characteristic curve for a soil is defined
ratory process. For most practical problems, it has been as the relationship between water content and suction for
found that approximate soil properties are adequate for the soil (Williams 1982). The water content defines the
analysis (Papagiannakis and Fredlund 1984). Hence, empir- amount of water contained within the pores of the soil. In soil
ical procedures to estimate in saturated soil parameters would science, volumetric water content 0 is most commonly used.
be valuable. In geotechnical engineering practice, gravimetric water con-
Laboratory studies have shown that there is a relationship tent w. which is the ratio of the mass of water to the mass of
between the soil-water characteristic curve for a particular soil solids, is most commonly used. The degree of saturation S
and the properties of the unsaturated soil (Fredlund and is another commonly used measure to indicate the percent-
Rahardjo 19930). For example, it has become an accept- age of the voids that are filled with water. The above vari-
able procedure to predict empirically the permeability fiinc- ables have also been used in a normalized form where the
tion for an unsaturated soil by using the saturated coeffi- water contents are referenced to a residual water content
cient of permeability and the soil-water characteristic curve (or to zero water content).
(Marshall 1958; Mualem 1986; University of Saskatchewan The suction may be either the matric suction (also known
1984). Similar procedures have been suggested for the shear as capillary pressure) of the soil (i.e., u , - u , , where ua is
strength properties of an ~ ~ n s a t u r a t esoil
d (Fredlund and the pore-air pressure and c i , is the pore-water pressure) or
Rahardjo 1993b). Since the soil-water characteristic curve is total suction (i.e., matric plus osmotic suction). At high suc-
~ ~ s as
e dthe basis for the prediction of other unsaturated soil
parameters, such as the permeability and shear-strength here are several soil terms that are used interchangeably in
the literature. The terminology used in the paper is most con-
fiinctions, it is important to have a reasonably accurate char- sistent with that found in the geotechnical literature. Other terms
acterization of the soil-water characteristic curve. are used in the geo-environmental,petroleum, and some of the soil
This paper reviews the forms of mathematical equations science disciplines. Some of these equivalences are as follows:
that have been s~iggestedto characterize the soil-water char- matric suction = capillary pressure, air-entry value = displace-
acteristic curve. It appears that none of the suggested equa- ment pressure, and soil-water characteristic curve = suction -
tions accurately fit laboratory data over the entire suction volumetic water content curve.
P l l i l L d lo C,to.ld.~ 1 lniprlinr. .la C.io.l&l
522 CAN. GEOTECH. J. VOL. 3 1. 1994
,
s i l t y soil 1 I
\
Can. Geotech. J. Downloaded from www.nrcresearchpress.com by Depository Services Program on 02/27/13
0.1 1 10 100 1000 10 000 100 000 1 000 000 0.1 1 10 100 1000 10000 100000 1000000
Matric Suction (kPa) Matric Suction (kPa)
FIG. 1. Typical soil-water characteristic curve for a silty soil. FIG.2. Soil-water characteristic curves for a sandy soil, a silty
soil, and a clayey soil.
tions (i.e., greater than about 1500 kPa), matric suction and
total suction can generally be assumed to be equivalent. Literature review
As a result of the different terminologies used, the soil- Numerous empirical equations have been proposed t o
water characteristic curves have taken on numerous forms. simulate the soil-water characteristic curve. Among the ear-
It is suggested that the term soil-water characteristic curve liest is an equation proposed by Brooks and Corey (1964).
be used to represent the relationship between volumetric It is in the form of a power-law relationship:
water content 0 and matric suction. Volumetric water content
test results in the low suction range are often presented
using an arithmetic scale. Soil-water characteristic curves
over the entire suction range are often plotted using a log-
For personal use only.
equation is not suitable in the high-suction range, since the is not suitable as a general fdrm, allhough it might apply
curve drops exponentially to zero at high suction values. for some soils over a limited range of suction values.
Equation [ l ] implies that there is a sharp discontinuity To establish a theoretical basis for the soil-water charac-
in suction near saturation. Although some coarse-grained teristic curve, let us consider the pore-size distribution curve
sands may have a rapid change in suction at low suctions, for the soil. The soil may be regarded as a set of intercon-
most soils, particularly medium- and fine-textured soils, nected pores that are randomly distributed. The pores are
show a gradual curvature in the air-entry region near satu- characterized by a pore radius r and described by a fimc-
ration. A modification of [ I ] was suggested by Roger and tion f(r), where f ( r ) d r is the relative volume of pores of
Hornberger (1978) to account for gradual air entry. In the case radius r to r + d r . In other words, f ( r ) is the density of
where the volumetric water content is referenced to zero pore volume corresponding to radius r. Since f(r) d r is the
water content and the normalized volumetric water content contribution of the pores of radius r to r + d r that are filled
@ (i.e., 810,) is plotted as the abscissa, the general soil-
Can. Geotech. J. Downloaded from www.nrcresearchpress.com by Depository Services Program on 02/27/13
metric water content and suction. S-shaped curve, and [20] can be used as a model to describe
( 3 ) Cnse where p o r e - s i z e function v a r i e s inversely this relationship.
a s r("'") - For the case of f(r) = Air""+", where rn is an The two fitting parameters (i.e., the mean value p, and
integer, the relationship between volumetric water content and the standard deviation o ) in 1201 are related to the air-entry
suction is value of the soil and the slope at the inflection point on the
soil-water characteristic curve. If the slope at the inflection
point is s and the air-entry value is +, then the standard
deviation o can be written as
[0, elsewhere
where and the soil-water characteristic curve defined by [ 181 can be
FREDLUND A N D XING 525
- Integration of the
Can. Geotech. J. Downloaded from www.nrcresearchpress.com by Depository Services Program on 02/27/13
FIG.3. A sample distribution using [29] and its integration FIG.4. A sample distribution using [30] and its integration.
(eq. [61).
written as
Note that [26] has the same form as [3], which was used
by McKee and Bumb (1984) to describe the soil-water char-
acteristic curve. Therefore, [3] gives the best results if the
pore-size distribution of the soil is c l o s e to a g a m m a
distribution.
(3) Case of a beta distribution
Consider the case of a beta distribution for the function
For personal use only.
f(r>:
0.1 1 to 100 1000 10000 1000001000000
Matric Suction (kPa)
In this case, the soil-water characteristic curve given by Figure 3 shows a sample probability distribution for [29]
[18] has greater flexibility. For a equal to P, [ l a ] gener- along with its integration (i.e., eq. [6]).It can be seen that the
ates a symmetrical S-shaped curve. For a greater than P, integration drops t o zero over a narrow suction range.
the curve is nonsymmetrical and has a higher air-entry value, Therefore, [6] is not suitable in the high-suction region.
a gentler slope near saturation, and a steeper slope near the Experimental data show that after the residual water con-
residual water content. For a less than P, the curve has a tent, the plot should decrease linearly to a value of about
smaller air-entry value, a steeper slope near saturation, and l o 6 kPa (Croney and Coleman 1961). To describe the soil-
a gentler slope near the residual water content. In the case water characteristic curve more accurately, the following
when a and p are integers, the soil-water characteristic distribution is suggested:
curve defined by [ l a ] and [27] is related to the binomial
probability function as follows (Mendenhall et al. 198 1):
Equation [30] and its integration form are shown in Fig. 4 for
the same set of parameters (i.e., a = llp, n, in). This dis-
tribution function drops more slowly than [29] as Jl increases
and, therefore, [30] produces a nonsymmetrical curve that is
closer to the experimental data.
Integrating [30] using [ l a ] gives the following relationship
between volumetric water content and suction:
Equation [28] has a form similar to that of [5] suggested r 1I n
by Roger and Hornberger (1978), which was used to account
for a gradual air entry. Note that defining r over the interval
[O, 11 does not restrict its use. The beta density function
can be applied to any interval by translation and a change in Figures 5-7 show the effect of varying the three param-
the scale. eters a , tz, and in on the shape of the soil-water characteristic
526 CAN. GEOTECH. J. VOL. 31, 1994
Can. Geotech. J. Downloaded from www.nrcresearchpress.com by Depository Services Program on 02/27/13
0.1 1 10 100 1000 10000 1000001000000 0 200 400 600 800 1000 1200 1400 1600
Matric Suction (kPa) Matric Suction (kPa)
FIG.6. Sample plots of [31] with a = 100 and m = 1 (n varies). FIG. 9. A best-fit curve to the experimental data of a till
(S. Vanapalli, personal communication, 1993).
For personal use only.
and draw a tangent line through this point (Fig. 8). Let
s denote the slope of the tangent line. Then, the three param-
eters a, n, and m are determined as follows:
0
Slope --
-Wp WI
curve. From Fig. 5 it can be seen that when n and m are where 4,is the intercept of the tangent line and the matric-
fixed, parameter a (with a unit of kPa) is closely related to suction axis (Fig. 8).
the air-entry value. In general, the value for the parameter a Small values of m result in a moderate slope in the high-
would be higher than the air-entry value. However, for small suction range, and large values of n produces sharp corner
values of m, the air-entry value can be used for parameter a. near the air-entry value (see Fig. 9). Another example of a
Figure 6 indicates that parameter n controls the slope of best-fit curve to the experimental data for a silty loam from
the soil-water characteristic curve. The distribution given Brooks and Corey (1964) is shown in Fig.lO.
by [30] attains its maximum value approximately at the In [31], 0 becomes equal to 0, when the suction is zero,
value of a. Therefore, the point (a, 0(a)) can be used to and 0 becomes zero when suction goes to infinity. It is also
approximate the inflection point. Using this information, a possible to use the degree of saturation for curve fitting,
graphical estimation for the three parameters can be obtained since the degree of saturation varies from 0 to 1. Gravimetric
from the soil-water characteristic curve. First, locate the water content can be similarly normalized for curve-fitting
inflection point (+;, Oi) on the soil-water characteristic plot purpose. Three plots are shown for the same soil (i.e., silty
FREDLUND AND XING 527
,
o Experimental dat
1 10 100 1000 10000 100000 1000000 0.1 1 10 100 1000 10000 1000001000000
Matric Suction (kPa) Matric Suction (kPa)
FIG.12. A sample plot of [36]. FIG. 13. A best-fit curve to the experimental data of a till
using [31] (S. Vanapalli, personal communication, 1993).
loam) in Fig. 11, using different ways of representing the
water content of the soil (i.e., degree of saturation, volu- nificantly affected, since the correction function C(+) is
metric water content, and gravimetric water content). approximately equal to 1 at low suctions.
Experimental data have previously shown that the suc- Figure 13 shows a best-fit curve to the experimental data
tion of a soil reaches a maximum value of approximately obtained for a glacial till, using [31]. A best-fit curve to
lo6 kPa at zero water content. This upper limit can be built the same experimental data using [36] is shown in Fig. 14.
into [31] as follows: It can be seen that the modified equation (i.e., eq. [36]) fits
A the data better than [31]. The main difference is that the
curve is forced by C(+) to zero at a suction of lo6 kPa.
A graphical estimation of the four parameters a , n, rn,
where C(+) is a correction function defined as and +,, in [36] can be obtained from a semilog plot of the
ln(1++/+,> soil-water characteristic curve. First, determine the suction
C(*)= I ~ [ I + ( ~ o o o o o o / + , ) ~
corresponding to the residual water content by locating a +,
point where the curve starts to drop linearly in the high-
where +, is the suction corresponding to the residual water suction range (Fig. 15). Numerical results show that, in
content 0,. most cases, [36] gives a satisfactory approximation for +,
It can be seen that C(l 000 000) is equal to zero. Therefore, 1500 kPa. Its magnitude will generally be in the range
+
at the limiting point where = lo6 kPa, the water content 0 1500-3000 kPa. Figure 15 uses +,
= 3000 kPa for illustra-
calculated from [36] is zero. A sample plot for [36] is shown tion purposes. Next, locate the inflection point (+i, Oi) on
in Fig. 12. The curve at the low-suction range is not sig- the semilog plot and draw a tangent line through this point
528 C A N GEOTECH J VOL 31, 1991
i
Can. Geotech. J. Downloaded from www.nrcresearchpress.com by Depository Services Program on 02/27/13
FIG. 15. A sample plot for the graphical solution of the four FIG. 18. A best-fit curve to the experimental data of Kidd
parameters ( a , la, m, and JI,) in [36]. Creek tailings (N. Yang, personal communication, 1992).
where
0.8
...
0
C
.- -Best-fit curve
...5 0.6 The slope s of the tangent line can be calculated as follows:
.-
0
0
g 0.4
0)
0"
0.2 where $, is the intercept of the tangent line on the semi-
log plot and the matric-suction axis (Fig. 15).
I I I I A graphical estimation only gives approximate values for
0 1
0 4 8 12 16 20
the parameters. To obtain a closer fit to experimental data,
Matric Suction (kPa)
the three parameters (a, 11, and in) in [36] can be determined
using a least-squares method, if the measured data for 0
FIG. 16. A best-fit curve to the experimental data of a sand
(data from Moore 1939). and t) are available. The idea is to choose the three param-
eters such that the calculated values from [36] are as close
as possible to the measured values. Therefore, the following
(Fig. 15). Let s denote the slope of the tangent line on the objective function (i.e., sum of the squared deviations of
semilog plot. Then, the fitting parameters a, tz, and In can be the measured data from the calculated data) is minimized
determined as follows: with respect to the three parameters a, n , and In.
[37] a =+i
where:
O(a, n, tn) is the objective function,
M is the total number of measurements, and
Bi and $i are measured values.
FREDLUND A N D XING 529
Can. Geotech. J. Downloaded from www.nrcresearchpress.com by Depository Services Program on 02/27/13
0.1 1 10 100 1000 10000 1000001000000 0 50 100 150 200 250 300
Matric Suction (kPa) Matric Suction (kPa)
FIG. 19. A best-fit curve to the experimental data of a sand FIG.21. A best-fit curve to the experimental data of a silt
(Soil Laboratory data, University of Saskatchewan). (S. Huang, personal communication, 1993).
For personal use only.
0
0 5 10 15 20 25 30 35 40
Matric Suction (kPa)
0.1 1 10 100 1000 10000 1000001000000
FIG.20. A best-fit curve to the experimental data of a sand Matric Suction (kPa)
(University of Toronto data; University of Saskatchewan 1984). FIG. 22. A best-fit curve to the experimental data of a silt
(Soil Laboratory data, University of Saskatchewan).
This is a nonlinear minimization problem. A curve-fitting
utility CFvIEW was coded based on [36] and [41] using a
quasi-Newton method. The detailed nonlinear curve-fitting
algorithm is presented in the Appendix. Best-fit curves for
a tailings sand, a silt, and a clay are shown in Figs. 16-23.
An arithmetic scale has been used when the experimental
data in the high-suction range are not available. It can be
seen, from these results, that [36] can be used to fit the
experimental data reasonably well over the entire suction
range 0-10"~a.
Some applications require an estimation of the residual
water content. The following slightly different form of [3 11
can be used to estimate the residual water content 0,:
Here, 0, and 0, are treated as two additional parameters. FIG. 23. A best-fit curve to the experimental data of an ini-
The five parameters a , n, m , 0,, and 0, in [42] can be sys- tially slurried Regina clay (data from Fredlund 1964).
tematically identified through a best-fit analysis on experi-
mental data. acteristic curve is uniquely determined from the proposed
general equation.
Conclusions The analysis in this paper provides not only a theoreti-
General empirical equations have been proposed to describe cal basis for most of the empirical equations but also proposes
the soil-water characteristic curve. Each equation has its a new, more general equation to describe the soil-water
own limitations. A general form of the relationship between characteristic curve. Based on the proposed equation, a
water content and suction was developed based on the pore- curve-fitting utility, CFVIEW, was coded. It was found that
size distribution of the soil. If the pore-size distribution of the equation fits experimental data reasonably well over the
a soil can be obtained or predicted, then the soil-water char- entire suction range from 0 to lo6 kPa.
530 CAN. GEOTECH. J. VOL. 31, 1994
r )
\
Gardner, W.R. 1958. Some steady state solutions of the unsatu- Sadler, D.R. 1975. Numerical methods for nonlinear regression.
rated moisture flow equation with application to evaporation University of Queensland Press, St. Lucia, Queensland,
from a water-table. Soil Science, 85: 228-232. Australia.
Gardner, W.R., Hillel D., and Benyamini, Y. 1970a. Post irri- University of Saskatchewan. 1984. KCAL user's manual. A com-
gation movement of soil water. I. Redistribution. Water puter program for calculating unsaturated permeability. Depart-
Resources Research, 6: 85 1-861. ment of Civil Engineering, University of Saskatchewan,
Gardner, W.R., Hillel D., and Benyamini, Y. 1970b. Post irri- Saskatoon.
gation movement of soil water. IS. Simultaneous redistribu- van Genuchten, M.T. 1980. A closed-form equation for pre-
tion and evaporation. Water Resources Research, 6: 1148- 1153. dicting the hydraulic conductivity of unsaturated soils. Soil
Marshall, T.J. 1958. A relation between permeability and size Science Society of America Journal, 44: 892-898.
distribution of pores. Journal of Soil Science, 9: 1-8. Williams, P.J. 1982. The surface of the Earth, an introduction
McCuen, R.H., Rawls, W.J., and Brakensiek, D.L. 1981. Statistical to geotechnical science. Longman Inc., New York.
analyses of the Brook-Corey and the Green-Ampt parame- Williams, J., Prebble, R.E., Williams, W.T., and Hignett, C.T.
ters across soil textures. Water Resources Research, 17: 1983. The influence of texture, structure and clay mineralogy
1005-1013. on the soil moisture characteristic. Australian Journal of Soil
McKee, C.R., and Bumb, A.C. 1984. The importance of unsat- Research. 21: 15-32.
Let p = ( a , n, m) denote the unknown vector of the three parameters a , n, a n d m and suppose that mea-
sured data (0,, ( J ~(i) = l , 2, ..., M ) a r e available, w h e r e M is t h e n u m b e r of measurements. T h e least-
squares estimate of p is the vector p:" which minimizes the following objective function (i.e., s u m of the
squared deviations of the measured data from the calculated data).
M
[A21 0 ( p ) = 0 ( a , m , n ) = ~ [ 8 ~ - 0 ( $ ~a , m , n ) 1 2
i=l
In other words, the least-squares method determines the three parameters such that the calculated values
from [ A l l a r e as close a s possible to the measured values.
A standard requirement of iterative minimization algorithms is that the value of the objective function
decreases monotonically from iteration to iteration. Let pi b e the estimate of p at the beginning of the ith iter-
ation ( p o is t h e initial guess and, theoretically, it is arbitrary). T h e new estimate pi+, is chosen such that
FREDLUND AND XING 53 1
5
O(p,,,) < O(p,).The steepest descent method is one of the easiest methods for mirhnizing' a general non-
linear function of several variables. It exploits the fact that from a given starting point a function decreases
most rapidly in the direction of the negative gradient vector evaluated at the starting point. Let g denote the
gradient of O ( p ) at p,. That is,
Can. Geotech. J. Downloaded from www.nrcresearchpress.com by Depository Services Program on 02/27/13
where a is a scalar that determines the length of the step taken in the direction of -g.
From [A21 it follows that
Similarly,
For personal use only.
The steepest descent method is not efficient for practical use, since the rate of convergence is slow, espe-
cially near the stationary point. The following quasi-Newton method (Sadler 1975) was used for the curve-
fitting program:
where:
g, is the gradient of the objective function evaluated at pi, and
A , is the operative matrix at the ith iteration.
Equation [A11 ] becomes the steepest descent method if Ai is the identity matrix multiplied by a step length
(a scalar). Denote pi+, - pi by di and g,,, - g, by q,. Then A i is updated using the following formula:
where:
a i is the ith element of the starting vector p,,
p, is the ith element of the gradient go evaluated at the starting vector.
The quasi-Newton method does not require matrix inversion or equivalent, since the sequence A i (i = 0,
C A N GEOTECH J VOL 31, 1994
b
1, 2, ...) converges to the inverse Hessian. In practice, the objective function is oftkn app;oximately quadratic
near the minimum, so a second-order convergence can be eventually expected. However, th&e is no guar-
antee that A, remain positive definite, even for a quadratic function. The product g?d, should be checked and
d l replaced by its negative, if glTd1> 0. Numerical difficulties may also arise when the scalar product (dl -
A,qJTq,is very small, resulting in unduly large elements in A , , , . One of several possible strategies is
to reinitialize A , , , if the cosine of the angle between ( d l - A,q,) and q , is less than 0.0001. For a non-
quadratic objective function it is reasonable to adjust the step length so that the objective function is
reduced at each iteration.
Can. Geotech. J. Downloaded from www.nrcresearchpress.com by Depository Services Program on 02/27/13
For personal use only.