You are on page 1of 11

Transportation Geotechnics 22 (2020) 100295

Contents lists available at ScienceDirect

Transportation Geotechnics
journal homepage: www.elsevier.com/locate/trgeo

Prediction of soil-water characteristic curves using basic soil properties T


Jinchun Chai, Peerapat Khaimook
Faculty of Science and Engineering, Saga University, Hojyo 1, Saga 840-8502, Japan

A B S T R A C T

A model for predicting the soil–water characteristic curve (SWCC) using easy-to-measure basic soil properties has been proposed. The model uses the widely used
SWCC function by Fredlund and Xing. This function has three parameters: a, n, and m. a is related to the air-entry value (ψaev ) of a soil, n mainly influences the
desaturation rate in the transition zone of a SWCC, and m controls the residual moisture content. By analyzing the results of a series of laboratory SWCC tests as well
as the results in the literature, explicit empirical equations were established for evaluating the values of a, n, and m using the saturated permeability (ks), grain size
distribution (GSD), and plasticity index (PI). The basic soil properties were selected based on the conceptual relationship between each SWCC function’s parameters
and those basic soil properties. The model has been verified by applying it to nine SWCCs in the literature, which were not included in developing the model. The
proposed model showed better predictions than the existing prediction models. It is suggested that the model can be used to estimate the preliminary drying SWCC of
soils.

Introduction an increase in the matric suction is called the rate of desaturation (δS ).
For modelling the mechanical behaviors of unsaturated soils, many
Unsaturated soil is commonly encountered in many geotechnical researchers [7,41,14,24] have developed equations to fit a tested
structures, such as dams, embankments and roads. Saturated soil me- SWCC, which is usually referred to as the best-fitted SWCC [26]. The
chanics are not applicable to unsaturated soils because the properties of SWCC equations are relationships between soil moisture contents and
the unsaturated soil, e.g., permeability, shear strength, etc., are a suctions. In recent studies, it was found that SWCC is not only depen-
function of the soil moisture content [13]. Natural disasters, such as dent on soil suction but also volume-mass properties. Therefore, SWCC
landslides or slope failures of road embankments during or after a is stress-dependent [29,40,48,10]. There are models considering the
rainfall, are mainly due to changes in unsaturated soil properties. For effect of void ratio change during SWCC test process [16,38,50], and a
example, the shear strength of soils generally decreases as their model for bimodal SWCC [44]. Comparing with Fredlund and Xing’s
moisture content increases [12]. The soil–water characteristic curve [14] model, these models have more fitting parameters, and Fredlund
(SWCC) describes a relationship between the soil moisture content and and Xing’s model is still a widely used one because its simplicity and its
matric suction, which is a fundamental foundation in unsaturated soil ability to fit with a large amount of test data. The equation is as follows:
mechanics [25,35,27,20,21,45]. Conventionally SWCC of a soil is di-
rectly measured using tensiometer, axis-translation method, etc, which ⎡
w (ψ) = ⎢1 −
(
ln 1 + C
ψ
r ) ⎤
⎥× ws
could take weeks or months to complete a measurement because of a m

considerably long equilibrium time under each suction. Recently, new



⎣ (
ln 1 +
1, 000, 000
Cr ) ⎥
⎦ {
ln ⎡exp +
⎣ ( ) ⎤⎦ }
ψ n
a (1)
techniques such as evaporation method using HYPROP, dewpoint po-
where w (ψ) is moisture content; ψ is soil matric suction; ws is saturated
tentiometer method using WP4C and small-scale centrifuge method are
moisture content; Cr is a correction factor related to the soil suction
available for obtaining SWCC of soils in a relatively shorter time-period
corresponding to the residual moisture content of the soil; a, m, and n
[28,8,23]. Alternatively, SWCC of a soil can also be obtained by using a
are fitting parameters. a is mainly related to the air-entry value (ψaev ), n
prediction model (e.g. [30].
is related to the rate of desaturation (δS ), and m is mainly related to the
As shown in Fig. 1, a SWCC can be divided into three zones, i.e.,
residual degree of saturation on SWCC (Sr).
saturated, transition, and residual zones. The matric suction separates
In most engineering practices where time and/or resource for de-
the saturated and transition zones is named the air-entry value (ψaev ).
termining unsaturated soil properties is limited, and a prediction ap-
The matric suction separates the transition and residual zone and is
proach is desirable. In the literature, several prediction models for the
called the residual suction (ψr ). At the residual suction, the corre-
SWCC have been proposed. The models can be categorized into two
sponding moisture content is called the residual degree of saturation
groups according to their predicting approaches, which are the physico-
(Sr). In the transition zone, the rate of moisture content reduction with

E-mail address: chai@cc.saga-u.ac.jp (J. Chai).

https://doi.org/10.1016/j.trgeo.2019.100295
Received 18 June 2019; Received in revised form 22 September 2019; Accepted 29 October 2019
Available online 01 November 2019
2214-3912/ © 2019 Elsevier Ltd. All rights reserved.
J. Chai and P. Khaimook Transportation Geotechnics 22 (2020) 100295

empirical model [4,15,36] and the functional parameter regression


model [46,30,43,22,9,11,34,33,49].
The concept of the physico-empirical model is to use basic soil
physical information (e.g., grain size distribution (GSD) and void ratio)
to estimate the pore size distribution, which is subsequently used to
calculate soil moisture contents and equivalent soil suctions using the
capillary theory [39]. Arya and Paris [4] proposed that the GSD of soils
with a similar shape to the SWCC can be divided into N subsections, and
each subsection corresponds to one point of matric suction and volu-
metric water content on a SWCC. Simms and Yanful [36] proposed a
similar theory but instead of assuming pore space/structure from GSD,
a SWCC was derived from measured pore size distribution and an ac-
curate prediction was shown. However, the model requires pore size
distribution which can be measured only by using a special equipment
(such as mercury intrusion porosimeter).
A functional parameter regression model assumes that parameters
of a SWCC equation can be correlated to the soil properties. Perera et al.
[30] modified Zapata's [46] model. For plastic soils (plasticity index
(PI) larger than 8), the percentage of weight passing the No. 200 sieve
Fig. 1. Zones in the SWCC and characteristic values.
(opening 0.075 mm), so-called the fine fraction (FF), and PI were cor-
related with the parameters of Fredlund and Xing's [14] SWCC equa-
tion. The equations are as follows:
a = 32.835 × ln(FF ·PI ) + 32.438(kPa) (2)

n = 1.421 × (FF ·PI )−0.3185 (3)

m = −0.2154 × ln(FF ·PI ) + 0.7145 (4)

where FF is in decimal form.


Cr = 500(kPa) (5)

For nonplastic soils (PI < 8), the parameters of the SWCC equation
can be calculated from the GSD data as follows:
a = 1.14α − 0.5, a ⩾ 1(kPa) (6)

where
α = −2.79 − 14.1 log(D20 ) − 1.9 × 10−6FF 4.34 + 7 log(D30) + 0.055Df
(7)
Fig. 2. Photo of the modified oedometer equipment: (a) soil chamber, (b) set- 40
tlement dial gauge, (c) vertical loading pressure, (d) vertical pressure and air
Df = 10⎡⎣ s1 + log(D60)⎤⎦ (8)
pressure gauges/regulator, (e) burette, and (f) bubble flooding tube.
30
s1 =
[log(D90) − log(D60 )] (9)

Fig. 3. Schematic diagram of the modified oedometer equipment.

2
J. Chai and P. Khaimook Transportation Geotechnics 22 (2020) 100295

Table 1
Basic soil properties of the soil samples.
CODE Gs Void ratio Atterberg's limits Grain size distribution analysis

PI PL (%) LL (%) D10 (mm) D15 (mm) D30 (mm) D60 (mm) Clay (%) Silt (%) Sand (%)

M#850 2.62 0.905 – – 21.55 – 0.0176 0.115 0.3229 11.68 12.03 76.29
M#425 2.66 0.887 – – 21.55 – 0.0088 0.0758 0.1715 13.68 16.07 70.25
M#250 2.65 1.043 – – 21.55 – 0.0103 0.059 0.1299 13.11 22.87 64.02
M#106 2.60 1.046 7.94 24.4 32.34 – 0.0035 0.0246 0.0512 15.88 67.68 16.44
M100 2.67 0.732 – – 21.55 0.0203 0.0932 0.2526 0.7102 7.29 6.13 86.58
M95C5 2.65 0.635 8.13 23.3 31.43 – 0.0038 0.1194 0.5404 16.07 9 74.93
M90C10 2.56 0.628 18.78 21.46 40.24 – – 0.029 0.4441 24.88 6.28 68.84
M80C20 2.65 0.908 19.26 23.56 42.82 – – 0.005 0.4353 29.98 4.9 65.12
S80C20 2.58 0.659 12.77 19.82 32.59 0.0047 0.0494 0.2203 0.7254 10.2 9.82 79.98
S70C30 2.56 0.679 12.41 16.84 29.25 – 0.0049 0.0926 0.5813 15.07 13.99 70.94
S60C40 2.55 0.971 14.01 26.48 40.49 – 0.0023 0.0467 0.5209 17.51 21.49 61

Table 2
Basic soil properties and best-fitted SWCC parameters of the experimental datasets.
CODE Class of data Soil type Basic properties for correlation Best-fitted SWCC

ks (m/s) δG FF∙PI FF/D15 (%/mm) a (kPa) n m R2

M#850 Type-1 NP 8.50E-07 2.23 – 1347.16 4.24 2.59 0.68 1.00


M#425 Type-1 NP 4.84E-07 2.82 – 3380.68 11.25 3.36 0.60 1.00
M#250 Type-1 NP 7.85E-07 2.92 – 3493.16 10.13 3.99 0.58 0.99
M#106 Type-1 NP 4.16E-07 3.14 – 23874.29 22.32 4.84 0.50 1.00
M100 Type-1 NP 1.43E-06 2.23 – 143.99 5.28 2.30 0.83 0.94
M95C5 Type-2 P 5.66E-08 1.56 203.80 – 60.77 1.00 2.17 0.91
M90C10 Type-2 P 1.46E-08 0.84 585.20 – 490.90 0.70 2.10 0.92
M80C20 Type-2 P 6.44E-09 0.49 674.10 – 1887.07 0.52 2.08 0.97
S80C20 Type-2 P 6.59E-08 1.93 255.70 – 82.14 0.92 2.16 0.95
S70C30 Type-2 P 1.86E-08 1.25 360.6 – 258.45 1.23 2.14 0.93
S60C40 Type-2 P 1.04E-08 1.15 546.4 – 721.02 1.18 2.10 0.93

NP: nonplastic soil, P: plastic soil, FF in %.

n = 0.936β − 3.8 (10) et al. [30] and Arya and Paris [4] is that some proposed correlations
lack physical or rational links between the basic soil parameters used
D and the parameters in a SWCC function. Relating all three parameters,
β = ⎧5.39 − 0.29 ln ⎡FF ⎛ 90 ⎞ ⎤ + 3Ds 0.57 + 0.021FF1.19⎫ s10.1
⎜ ⎟

⎨ ⎢ ⎝ D10 ⎠ ⎥ ⎬ a, n and m in Fredlund and Xing’s [14] SWCC function to plasticity


⎩ ⎣ ⎦ ⎭ (11)
index (PI) and fraction of fine (FF) [30] lacks physical or theoretical
−30 + log(D ) bases; and dividing soil particles into several sub-sections and each
Ds = 10⎡⎣ s2 30 ⎤
⎦ (12)
section has soil particles with a single size (diameter) [4] is far away
20 from the reality.
s2 =
[log(D30) − log(D10 )] (13) Based on the above discussions, this study tries to establish a model
for predicting SWCC using basic soil properties under physical analogy
m= 0.26e0.758χ + 1.4D10 (14) or rational links between the basic soil parameters and the parameters
where D10, D20, D30, D60, D90 are diameters of soil particles corre- in a SWCC function. For examples, saturated permeability to the
sponding to 10, 20, 30, 60 and 90% passing on GSD, and parameter, a (related to air-entry value), slope of the middle part of
GSD to the parameter, n (related to the rate of desaturation on SWCC),
1 etc. Then a series of SWCC tests were conducted, and SWCC data from
χ = log s21.15 − ⎛1 − ⎞
⎝ n⎠ (15) literatures were gathered to form a database for regression analysis. The
mathematical links between parameters in SWCC function and the se-
Cr = 100(kPa) (16)
lected basic soil properties were established by regression analyses to
Sahin et al. [34] and Zhang et al. [49] also proposed a set of re- form a new functional parameter regression model for the SWCC. The
gression equations by correlating all parameters of SWCC equation with performance of the model is compared with the existing models.
methylene blue value (MBV) which is generally employed in the oil
industry field. Laboratory tests on the SWCC
Another approach for predicting SWCC of a soil using basic soil
properties and one-point (or two points) measurements on SWCC The laboratory tests were conducted in a modified oedometer by
[43,22,9,11,33]. Houston et al. [22] proposed that Perera et al.’s [30] employing the axis-translation technique and the testing procedures
model could be improved by adjusting predicted SWCC to converge follows the ASTM D-6836-16 standard [5]. Twelve (12) soils tested
with one measured data point on SWCC. Similarly, Chin et al. [11] were generated by original three (3) soils, namely a marine clay (Ariake
proposed a set of equations for predicting values of parameters in clay), a decomposed granite (Masado, d < 2.0 mm) and a fine sand
Fredlund and Xing’s [14] equation using D50 on GSD curve and mea- (d < 2.0 mm) with different mixing proportions. Sample codes were
sured two data points (corresponding to suctions 10 kPa and 500 kPa) designed to indicate the amount of each component in the mixture. In
on SWCC. While it is not a pure prediction model. the sample codes, “M” means the Masado; “S” for the sand; “C” for the
The main limitation of the prediction models propose by Perera clay; the numbers followed these letters mean percentage amount of the

3
J. Chai and P. Khaimook Transportation Geotechnics 22 (2020) 100295

Fig. 5. m – FF∙PI relationship.

Fig. 4. Grain size distribution of tested soils: (a) sieved Masado soils, (b)
Masado-clay mixtures, (c) sand-clay mixtures. Fig. 6. m − FF (D15 D0 ) relationship.

corresponding component; and “#” indicate that the number followed their liquid limits and were de-aired in a vacuum chamber for at
it is the sieve number. For example, sample M80C20 contains 80% least 24 hrs. with a vacuum pressure of 80 kPa. The ceramic disk
Masado and 20% Ariake clay by dry weight; and M#850 means the was also saturated by inundation it under the water in the vacuum
Masado sieved with the #850 sieve (with opening size of 0.85 mm). The chamber. The disk used had an air-entry value of 500 kPa.
photo and schematic diagram of the equipment are shown in Figs. 2 and (2) The water was properly added to the burette to set the initial water
3, respectively. The test procedures are as follows: head at the base level of the soil sample, and the tube system was
flooded to ensure that there were no air bubbles in the system.
(1) Sample and ceramic disk preparation: The soil samples were pre- (3) The ceramic disk was installed into the base of the modified
pared from slurries whose initial water content were more than

Table 3
Basic soil properties and best-fitted SWCC parameters of the literature datasets.
CODE Class of data Soil-type Basic properties for correlation Best-fitted SWCC Source

2
ks (m/s) δG FF∙PI FF/D15 (%/mm) a (kPa) n m R

1464 Type-1 NP 2.31E-05 3.945 – 77.23 3.66 5.34 0.95 0.99 (United States Department of [1]
1466 Type-1 NP 2.48E-05 6.106 – 393.19 5.68 7.34 0.90 1.00
2220 Type-1 NP 6.99E-06 2.357 – 632.08 6.49 3.50 0.82 0.98
2221 Type-1 NP 1.45E-04 1.668 – 332.80 4.73 2.50 1.02 0.99
BBA2 Type-1 P 8.28E-08 – 11.20 – 18.317 1.13 1.86 1.00 [32]
NTU-3A Type-1 P 3.05E-09 – 16.08 – 1397.81 0.46 1.85 0.94 [2,3]
SLR-2B Type-1 P 3.00E-08 – 42.16 – 378.33 0.62 1.40 0.98
SLR-2A Type-1 P 2.78E-08 – 35.96 – 468.22 0.54 1.44 0.94
SLR-2C Type-1 P 1.36E-08 – 14.25 – 136.68 0.57 1.95 1.00
BN1 Type-1 P 1.40E-08 – 6.25 – 506.22 0.63 2.17 0.99

NP: nonplastic soil, P: plastic soil, FF in %.

4
J. Chai and P. Khaimook Transportation Geotechnics 22 (2020) 100295

(7) The air pressure and vertical pressure were increased for the next
matric suction, where the effective vertical pressure of 50 kPa was
maintained.
(8) Steps 5 and 6 were repeated until the final suction was applied.
(9) The soil sample was taken from the chamber, and the final water
content was measured. The degree of saturation corresponding to
each matric suction was back-calculated, and a main drying SWCC
was obtained.

The basic properties of the tested soils are listed in Table 1 (the
saturated permeability is given in Table 2), and their GSDs are shown in
Fig. 4.

Data collection from the literature and data interpretation

Data collected from the literature

Nineteen published SWCC data from eight different sources


[32,31,2,3,1,19,17,18] were collected from the literature. Ten pub-
Fig. 7. n − δG relationship. lished data were used to develop a new prediction model, and another
nine datasets were used to verify the model.

Data interpretation

Grouping the datasets


For a soil containing high percentages of silt and clay, there is
usually difficulty in measuring the SWCC under residual conditions
because of the limitation of the maximum suction pressure that can be
achieved in a laboratory. Zhai and Rahardjo [47] reported that the
uncertainty of a so-called best-fitted SWCC increased as the maximum
applied matric suction decreased and as the number of measured data
points decreased. In this study, the SWCC data, which contain at least 4
data points and have a range of degree of saturation covers from more
than 85% to below 30%, are classified as relatively complete SWCCs
and named Type-1 datasets and others as Type-2 datasets.

A method of fitting the SWCC


According to Fredlund et al. [15], the correction factor (Cr ) in Eq.
(1) can be selected to be 3000 kPa and the saturated degree of satura-
Fig. 8. a – ks/k0 relationship.
tion (ws ) equals 100%. Then, only a , n , and m need to be fitted. The
determination process of the fitted parameters employs the least square
method [37]and uses the Solver function of Microsoft Excel to make a
oedometer chamber, which has a diameter of 60 mm and thickness number of tries to find the best set of SWCC parameters in Eq. (1). The
of 20 mm. A sealant was applied between the ceramic disk and the configuration of the trial process is as follows:
chamber. A filter paper was placed on the ceramic disk before
filling the soil sample to prevent fine soil particles entering the disk.
A collar with an inner diameter of 60 mm was set on the top of the (1) Range 10−1 ⩽ a ⩽ 105 (kPa)
oedometer ring. A saturated slurry sample was then filled into the 0.001 ⩽ n ⩽ 20
0.001 ⩽ m ⩽ 4
chamber and the collar with an initial thickness of 35 mm and
(2) Increment Increment for a power value of 0.001 for a
preconsolidated under a vertical pressure of 50 kPa for 24 h. 0.001 for n
(4) After preconsolidation, the top part of the sample was trimmed to 0.001 for m
make the initial height of the specimen 20 mm. Then, the axis (3) Initial value 101 for a (kPa)
translation technique was used to measure the suction and moisture 1 for n
content relationship. To ensure a firm contact between the ceramic 1 for m

desk and a soil specimen, an effective confining (consolidation) The value of m in Eq. (1) related to the residual state of a SWCC, and in
stress is needed and a value of 50 kPa was adopted which is establishing a model to predict it, only the Type-1 datasets, which have
equivalent to an overburden pressure of 4–5 m of soil. relatively complete SWCCs, were used. For Type-2 datasets, in the fit-
(5) Under a given matric suction, the amount of water in the burette ting process, the value of m was predetermined using the established
and the settlement of the soil sample were recorded periodically model first, and only a and n were fitted.
until there was no change in the water level in the burette for 24
hrs. This was then considered that the equilibrium state was A new model for predicting the SWCC
achieved. The current water level in the burette and settlement
were measured, which were used to calculate the degree of sa- Basic consideration for selecting the basic soil properties
turation corresponding to the current matric suction.
(6) During the test, the water was flooded into the tube systems every As discussed previously that parameters a, n, m in Fredlund and
72 h. to remove possible dissolved air bubbles. Xing’s [14] SWCC function are mainly related to air entry value (ψaev ),

5
J. Chai and P. Khaimook Transportation Geotechnics 22 (2020) 100295

Fig. 9. Prediction of the SWCC using tested data of plastic soils: (a) S60C40, (b) S70C30, (c) S80C20, (d) M80C20, (e) M90C10, and (f) M95C5.

rate of desaturation (δS ), and residual degree of saturation on SWCC (Sr) ψaev , which is related to the parameter a.
respectively. It is considered that the easy to be measured basic soil (2) δS : A well-graded soil potentially has a pore structure in which the
properties used for predicting SWCC of a soil should have links with the radius of the pores has a wide range and a slower desaturation rate
physical meaning of each parameter of a SWCC. on its SWCC than a soil with a single-dominant pore size (i.e., a
poorly graded soil). Soil gradation can be determined using a
(1) ψaev : Ideally, ψaev can be linked with the largest continuous pore size coefficient of uniformity. However, there are cases in which soils
of a soil [4,15], but it is unlikely to be measured in general en- have higher silt and/or clay contents, and the diameter of 10%
gineering practice. On the other hand, saturated permeability (ks ), passing (D10) cannot be measured by a hydrometer test. In this
which describes the flow capacity of a soil for water, is relatively study, a parameter (δG ) derived from GSD curve is introduced. First,
easier to measure. The flow of water is analogous to the flow of air the GSD slope parameter (δG1) is defined as follows:
in a soil. Therefore, it is considered that ks can be correlated with

6
J. Chai and P. Khaimook Transportation Geotechnics 22 (2020) 100295

Fig. 10. Prediction of the SWCC using experimental data: (a) M100, (b) M#850, (c) M#425, (d) M#250, and (e) M#106.

δG1 =
60 − 30
=
30 are two main factors influencing Sr. One is the fine fraction
log(D60) − log(D30) log( D60 D30 ) (17) (< 0.075 mm) (FF), and the higher the FF is, the higher the Sr.
Another factor is the type of clay mineral. Different clay minerals
In Eq. (17), 30 is a constant; for simplicity, replacing 30 in Eq. (17)
have different water absorption capacity. For example, montmor-
by 1, parameter δG is defined as:
illonite has the highest water absorption capacity. The plasticity
−1
index (PI) can be an indirect measurement of the mineral type of a
⎡ ⎤
δG = ⎢log ⎜⎛ D60 D30 ⎟⎞⎥ clay. Therefore, it is considered to relate the FF∙PI with Sr and
⎣ ⎝ ⎠⎦ (18) therefore parameter m in Eq. (1). Zapata [46] and Perera et al. [30]
It is considered that δG can be related to δS of a SWCC and therefore also used the FF∙PI to estimate parameter m. In the case of non-
to parameter n in Eq. (1). plastic soils (PI < 8) with FF > 5%, FF was also used. In addition,
to consider the effect of the particle size of the finer portion of a
(3) Sr: Sr is conceptually explained as a retained film of water in the soil, D15 was used. A soil with a larger value of D15 has a smaller
pore structure that is attracted to the surface of soil particles by the surface area in the microstructure, which in turn results in less
adhesion force [42]. For plastic soils (PI ≥ 8), conceptually, there water film attraction on the surfaces of particles and therefore a

7
J. Chai and P. Khaimook Transportation Geotechnics 22 (2020) 100295

−0.119
Table 4 FF ⎞
ARE of the predictions on tested data. m = 1.67 × ⎛ ⎜ ⎟

⎝ D15 D0 ⎠ (20)
Type of soils Tested soils Absolute relative error (ARE)
data
where FF is in %.
Proposed Perera et al. Arya and Paris
model [30] [4] n − δG relationship
Fig. 7 shows the regression results of the n − δG relationship. In this
Plastic S60C40 0.150 0.072 0.670
S70C30 0.126 0.244 0.590
analysis, the grain-size distribution data of the plastic soils from the
S80C20 0.242 0.431 0.648 literature are not available; therefore, they were not included. A re-
M80C20 0.140 0.048 0.582 gression analysis using both plastic and nonplastic soils resulted in an
M90C10 0.114 0.154 0.610 R2 value of 0.86. The equation is as follows:
M95C5 0.216 0.490 0.489
Average 0.165 0.240 0.598 n = 0.952δG1.185 (21)
Nonplastic M100 0.256 – 0.898
M#850 0.371 – 0.171 It can be observed that there are some scattered data. The reasons
M#425 0.029 – 0.178 considered are as follows: (1) δG cannot express the uniformity of a soil
M#250 0.106 – 0.177 completely, and (2) the slope of a SWCC in the transition zone might
M#106 0.068 – 0.272
Average 0.166 – 0.339
not be purely controlled by the parameter n in Eq. (1).

a – ks/k0 relationship
Table 5 The result of the regression analysis of the a – ks/k0 relationship is
Basic soil properties and predicted SWCC parameters of the literature datasets shown in Fig. 8. A unit permeability, k0 = 1 m/s, was introduced to
for verification. normalize ks. The values of a and ks/k0 have an inverse proportional
No. CODE Soil type Basic soil properties Source
relationship, i.e., a soil with a low saturated permeability would have a
high ψaev and therefore a large value of a. The regression analysis results
ks (m/s) δG FF∙PI FF/D15 in an R2 of 0.86, and the regression equation is as follows:
(%/mm)
−0.184
0.198 ⎛ ks k 0 ⎞
1 SK-10 P 7.67E-09 1.76 12.09 – [17] a = exp ⎝ ⎠ (ain kPa) (22)
2 SK-17 P 1.63E-08 0.81 5.90 – [18]
3 HR P 9.20E-07 0.80 16.00 – [31] where k0 = 1 m/s.
4 BN1 P 1.40E-08 1.11 19.21 – [32]
5 1010 NP 1.98E-05 4.08 – 724.05 (United States Verification of the proposed model
6 1160 NP 6.46E-05 1.99 – 2,129.20 Department of
7 1465 NP 4.63E-06 4.39 – 814.74 [1]
8 1467 NP 1.27E-06 1.16 – 111.62
The predicting model was used to predict the SWCCs for two da-
9 3311 NP 3.34E-05 1.58 – 12,680 tasets. The first dataset included 11 SWCCs of experimental data that
were used in establishing the regression relationships (Eqs. (19(–(22)),
NP: nonplastic soil, P: plastic soil, FF in %. and the second dataset includes 9 datasets from the literature that were
not used in establishing the proposed model. The performance of the
smaller value of Sr. We used D15 instead of D10 because for some of proposed model was evaluated by comparing predictions with the
the tested soils the values of D10 were not available. Therefore, for measured results as well as with the predicted results using the models
nonplastic soils, it is considered that parameter m can be related to proposed by Arya and Paris [4] and Perera et al. [30]. To evaluate the
FF/D15. model’s performance, the absolute relative error (ARE) has been in-
troduced as an indicator and can be calculated as follows:
Regression analyses k
|yi − yi | ⎞
1
A RE =
k
∑⎛ ⎜

y

The regression analyses were conducted using the tested and lit- i=1 ⎝ i ⎠ (23)
erature data, which are summarized in Tables 2 and 3, respectively.
Comparison of predicted and measured SWCCs - laboratory test datasets
m - FF∙PI relationship for plastic soils
Regression analysis of m - FF∙PI relationship was conducted using The predictions of the SWCC of 11 experimental data are shown in
the Type-1 datasets of plastic soils. The relationship between m - FF∙PI is Figs. 9 and 10 for plastic soils and nonplastic soils, respectively. The
shown in Fig. 5 with a high coefficient of determination (R2) of 0.95 values of the ARE are listed in Table 4. For plastic soils, it can be seen
and the equation is as follows: that in most cases, the SWCCs predicted by the proposed model agree
well with the experimental data. Fig. 9a and 9d show relatively poorer
m = [(−2.04 × 10−4) × (FF ·PI )] + 2.22 (19) predictions than the others. It can be seen that the whole curves were
shifted toward lower values of the matric suction compared to the ex-
where FF is in %. Parameter m of each soil in the Type-2 datasets was
perimental data, which indicates an underestimation of parameter a.
calculated by Eq. (19) before fitting a and n. However, the values of m
Fig. 9c and f show underestimations of ψ in the transition zone and in
parameter of 6 experimental data from the Type-2 datasets were di-
the residual zone, which can be accounted for the overestimated value
rectly calculated and included in Fig. 5 to enlarge the range of the m -
of n and/or m. However, the performance of the proposed model is still
FF∙PI relationship.
comparatively fair. Furthermore, it can be observed that the accuracy of
the estimated a has more influence on the predicted SWCC than that of
FF
m − (D D ) relationship for nonplastic soils (FF > 5%) n or m. For nonplastic soils, the predictions of the SWCC (Fig. 10)
15 0 FF
For nonplastic soils, the regression analysis of m − (D D ) re- mostly performed well. Relatively, there is no obvious difference in the
15 0
lationship was conducted (D0 = 1 mm). The relationship between is performance of the proposed model for nonplastic soils and plastic soils
2
shown in Fig. 6, with an R of 0.89. The regression equation is as fol- (average ARE of 0.166 and 0.165, respectively).
lows: The predictions using the existing models are also included in

8
J. Chai and P. Khaimook Transportation Geotechnics 22 (2020) 100295

Fig. 11. Prediction of the SWCC using literature data: (a) SK-10, (b) SK-17, (c) HR, and (d) BN1.

Figs. 9 and 10. For nonplastic soils, Perera et al.’s [30] model requires the predictions and the measured data in the case of HR (Fig. 11c) and
D10, which is unavailable in this dataset, therefore, this model is not BN1 (Fig. 11d), the proposed model generally shows good performance.
applicable. The predictions of Arya and Paris’s [4] model significantly For the nonplastic soils (Fig. 12), the model shows good SWCC pre-
underestimated the SWCC in every case of the plastic soils, but the diction in every case except the case of 1467 in Fig. 12d. In comparison,
model performed well for the nonplastic soils. It was obvious that the the performance of the proposed model applied on the nonplastic soils
effect of the PI, which was neglected in Arya and Paris’s [4] model, is better than that of the plastic soils with an average ARE of 0.261 and
caused poor predictions of the SWCCs of the plastic soils. On the other 0.457, respectively.
hand, Perera et al.’s [30] model, which used the PI and FF to estimate For the plastic soils, Arya and Paris’s [4] model showed poor pre-
the SWCC, generally overestimated the SWCCs in most cases for the dictions for all the datasets. Perera et al.’s [30] model significantly
plastic soils, except for the case of M80C20 (Fig. 9d). The basic reason overestimated the Sr for most cases. For the plastic soils, the average
considered is the weak physical correlation of FF∙PI with ψaev . For plastic ARE of the proposed model, Perera’s [30] model and Arya and Paris’ [4]
soils, generally, the proposed model’s performance (average ARE of model are 0.457, 2.633 and 1.605, respectively, and 0.261, 0.460, and
0.165) is comparatively better than that of Perera et al.’s [30] model 0.317, respectively, for the nonplastic soils.
(average ARE of 0.240) and Arya and Paris’s [4] model (average ARE of Although the idea of using some basic soli properties to predict
0.598). For the nonplastic soils, the average ARE of the proposed model SWCC of a soil is not new. However, linking the parameters in a SWCC
and Arya and Paris’s [4] model are 0.166 and 0.339, respectively, function to basic soil properties through physical analogy and/or ra-
which indicates that the proposed model performed better than the tional correlation is new. Linking air-entry value to saturated perme-
existing models. ability and the rate of desaturation to a slope on the grain size dis-
tribution (GSD) curve of a soil is novel. It is suggested that the proposed
model can be used to establish preliminary first drying SWCC of a soil
Comparison of predicted and measured SWCCs - literature datasets using easy-to-measure basic soil properties, i.e., ks, δG , FF, and PI for
plastic soils and ks, δG , FF, and D15 for nonplastic soils (FF > 5%).
Nine soil datasets (Table 5) from the literature that were not in-
cluded in the development of the proposed model were used to verify
the proposed model. The prediction results are shown in Figs. 11 and 12 Summary and conclusions
together with the predictions by the existing models. The ARE values of
both the proposed and exiting models are listed in Table 6. For the Using a series of laboratory-measured soil-water characteristic
plastic soils (Fig. 11), although there are some discrepancies between curves (SWCCs) and several sets of SWCCs from the literature, a new

9
J. Chai and P. Khaimook Transportation Geotechnics 22 (2020) 100295

Fig. 12. Prediction of the SWCC using literature data: (a) 1010, (b) 1160, (c) 1465, (d) 1467, and (e) 3311.

SWCC prediction model using basic soil properties has been proposed. based on their common correlation to pore size. It is reasoned that
This model is based on the widely used SWCC function equation pro- the slope of the middle part of grain size distribution (GSD), defined
−1
posed by Fredlund and Xing [14]. The feature of the model and its
verification process are as follows:
( )
as δG = log D60 D30 , has a physical link to δs and then the para-
meter, n. Finally FF∙PI and FF/D15 were chosen for evaluating m for
plastic soils and nonplastic soils respectively based on the con-
(a) Data collection and analysis: Eleven laboratory-measured SWCCs ceptual link between specific surface area, amount of fine particle
and 10 SWCCs from the literature as well as their basic soil prop- and mineral type with Sr.
erties were used in this study. Fredlund and Xing’s SWCC function (c) Establishment of correlations: By regression analysis, as well as the
has three parameters, a, n, and m. For each SWCC, the corre- authors’ judgement and suitable explicit functions of a – ks/k0
sponding values of each parameter were determined by best-fitting FF
(k0 = 1 m/s), n − δG , m – FF∙PI, and m − D D (D0 = 1 mm) were
15 0
Fredlund and Xing’s function using the least square method. established, which form a new SWCC prediction model.
(b) Selecting easy-to-measure basic soil properties: Parameters, a, n, (d) Verification of the proposed model: The proposed model was ver-
and m, which mainly influence air entry value (ψaev ), desaturation ified by predicting the SWCCs of 11 laboratory-tested soils and 9
rate (δs ) and residual degree of saturation (Sr ) on a SWCC, respec- SWCC datasets from the literature, which were not used in devel-
tively. The saturated permeability (ks) was chosen for estimating a oping the model. Using, absolute relative error (ARE), it was shown

10
J. Chai and P. Khaimook Transportation Geotechnics 22 (2020) 100295

Table 6 drying-wetting cycles. Geotech Environ Eng 2014:140.


ARE of the predictions on the literature datasets. [19] Goh SG, Rahardjo H, Leong EC. Modification of triaxial apparatus for permeability
measurement of unsaturated soils. Soils Found 2015;55:63–73.
Type of soils Verification data Absolute relative error (ARE) [20] Han Z, Vanapalli SK, Zou WL. Integrated approaches for predicting soil-water
characteristic curve and resilient modulus of compacted fine-grained subgrade soils.
Proposed Perera et al. Arya and Can Geotech J 2016;54:646–63.
model [30] Paris [4] [21] Hou X, Vanapalli SK, Li T. Water infiltration characteristics in loess associated with
irrigation activities and its influence on the slope stability in Heifangtai loess
Plastic SK-10 0.069 0.078 0.194 highland, China. Eng Geol 2018;234:27–37.
SK-17 0.043 0.188 0.116 [22] Houston WN, Dye HB, Zapata CE, Perera YY, Harraz A. Determination of SWCC
using one point suction measurement and standard curves. Fourth International
HR 1.538 9.826 5.854
Conference on Unsaturated Soils, 2006, Arizona. 2006.
BN1 0.179 0.441 0.255
[23] Khanzode RM, Vanapalli SK, Fredlund DG. Measurement of soil-water characteristic
Average 0.457 2.633 1.605
curves for fine-grained soils using a small-scale centrifuge. Can Geotech J
Nonplastic 1010 0.199 0.580 0.332 2002;39(5):1209–17. https://doi.org/10.1139/t02-060.
1160 0.210 0.758 0.194 [24] Kosugi K. Lognormal distribution model for unsaturated soil hydraulic properties.
1465 0.223 0.346 0.694 Water Resour Res 1996;32(9):2697–703. https://doi.org/10.1029/96WR01776.
1467 0.572 0.297 0.286 [25] Leong EC, Rahardjo H. Permeability functions for unsaturated soils. J Geotech
3311 0.101 0.317 0.081 Geoenviron Eng 1997;123:1118–26.
Average 0.261 0.460 0.320 [26] Leong EC, Rahardjo H. Review of soil-water characteristic curve equations. J
Geotech Geoenviron Eng 1997;123:1106–17.
[27] Li X, Li JH, Zhang LM. Predicting bimodal soil-water characteristic curves and
permeability functions using physically based parameters. Comput Geotech
that for both the nonplastic (PI < 8) and plastic soils (PI ≥ 8), the 2014;57:85–96.
proposed model resulted in fair to good predictions compared with [28] Lu N, Likos WJ. Unsaturated soil mechanics. John Wiley & Sons Inc; 2004.
the measurements and better than the results of the existing models. [29] Ng CWW, Pang YW. Influence of stress state on soil-water characteristics and slope
stability. J Geotech Geoenviron Eng, ASCE 2000;126(2):157–66.
It is suggested that the proposed model can be used to estimate the [30] Perera YY, Zapata CE, Houston WN, Houston SL. Prediction of the soil-water
main drying SWCC of a soil with the FF > 5%. characteristic curve based on grain-size-distribution and index properties. Geotech
Spec Publ 2005;130:49–60.
[31] Rahardjo H, Ong TH, Rezaur RB, Leong EC, Fredlund DG. Response parameters for
Declaration of Competing Interest characterization of infiltration. Environ Earth Sci 2010;60:139–1380.
[32] Rahardjo H, Satyanaga A, Leong EC, Ng YS. Variability of residual soil properties.
The authors would like to declare that the afore mentioned manu- Eng Geol 2012;141–142:124–40.
[33] Rahimi A, Rahardjo H. A new approach to improve soil-water characteristic curve
script does not contain any material of financial or other biased per-
to reduce variation in estimation of unsaturated permeability function. Can Geotech
sonal interest of the authors. J 2016;53(4):717–25.
[34] Sahin H, Gu F, Lytton RL. Development of soil-water characteristic curve for flexible
base materials using the methylene blue test. J Mater Civ Eng 2015;27(5).
Appendix A. Supplementary material
[35] Schnellmann R, Rahardjo H, Schneider HR. Unsaturated strength of a silty sand. Eng
Geol 2013;162:88–96.
Supplementary data to this article can be found online at https:// [36] Simms PH, Yanful EK. Predicting soil-water characteristic curves of compacted
doi.org/10.1016/j.trgeo.2019.100295. plastic soils from measured pore-size distribution. Géotechnique
2002;52(4):269–78.
[37] Stigler SM. Gauss and the invention of least squares. Ann Statist 1981;9:465–74.
References [38] Tarantino A. A water retention model for deformable soils. Géotechnique
2009;59(9):751–62.
[39] Taylor DW. Fundamentals of soil mechanics. New York: John Wiley and Sons; 1948.
[1] U.S.D.A. 1999. UNSODA MODEL 2.0 [computer software]. Riverside; CA.
[40] Trinh MT, Rahardjo H, Leong EC. Soil-water characteristic curve and consolidation
[2] Agus SS, Leong EC, Rahardjo H. Soil-water characteristic curves of Singapore re-
behavior for a compact silt. Can Geotech J 2007;44:266–75.
sidual soils. Geotech Environ Eng 2001;19:285–309.
[41] van Genuchten MT. A closed form equation for predicting the hydraulic con-
[3] Agus SS, Leong EC, Rahardjo H. Estimating permeability function of Singapore re- ductivity of unsaturated soils. Soil Sci Soc Am J 1980;44.
sidual soils. Eng Geol 2005;78:119–33.
[42] Vanapalli SK, Sillers WS, Fredlund MD. The meaning and relevance of the residual
[4] Arya LM, Paris JF. A physico-empirical model to predict the soil moisture char-
state to unsaturated soils. 51st Canadian Geotechnical Conference, 1998, Alberta.
acteristic from particle-size distribution and bulk density data. Soil Sci Soc Am J
1998.
1981;45:1023–30.
[43] Vanapalli SK, Catana MC. Estimation of the soil-water characteristic curve of
[5] ASTM 2016. ASTM Standard D6836-16 Standard Test Methods for Determinaion of
coarse-grained soils using one point measurement and simple properties.
the Soil Water Characteristic Curve for Desorption Using Hanging Column, Pressure
International Symposium on Advanced Experimental Unsaturated Soil Mechanics,
Extractor, Chilled Mirror Hygrometer, or Centrifuge. West Conshohocken, PA: 2005, Trento. 2005.
ASTM International.
[44] Wijiya M, Leong EC. Equation for unimodal and bimodal soil–water characteristic
[7] Brooks RH, Corey AT. Hydraulic properties of porous media. Colorado State
curve. Soils Found 2016;56(2):291–300.
University (Fort Collins); 1964.
[45] Xie WL, Li P, Vanapalli SK, Wang JD. Prediction of the wetting-induced collapse
[8] Bulut R, Leong EC. Indirect measurement of suction. J Geotech Geol Eng
behaviour using the soil-water characteristic curve. J Asian Earth Sci
2008;26(6):633–44. 2018;151:259–68.
[9] Catana MC, Vanapalli SK, Garga VK. The water retention characteristics of com-
[46] Zapata CE. Uncertainty in soil-water-characteristic curve and impacts on un-
pacted clays. Fourth International Conference on Unsaturated Soils, 2006, Arizona.
saturated shear strength predictions PhD. Ph.D. Thesis Arizona State University;
2006.
1999.
[10] Chen Y. Soil-water retention curves derived as a function of soil dry density.
[47] Zhai Q, Rahardjo H. Quantification of uncertainties in soil-water characteristic
GeoHazards 2018;1:3–19.
curve associated with fitting parameters. Eng Geol 2013;163:144–52.
[11] Chin KB, Leong EC, Rahardjo H. A simplified method to estimate soil-water char-
[48] Zhao Y, Cui Y, Zhou H, Feng X, Haung Z. Effects of void ratio and grain size dis-
acteristic curve. Can Geotech J 2010;47(12):1382–400. tribution on water retention properties of compacted infilled joint soils. Soil Found
[12] Fredlund DG, Rahardjo H. Soil Mechanics for Unsaturated Soils. New York: John
2017;57:50–9.
Wiley & Sons; 1993.
[49] Zhang J, Peng J, Chen Y, Li J, Li F. Estimation of soil-water characteristic curve for
[13] Fredlund DG, Rahardjo H, Fredlund MD. Unsaturated soil mechanics in engineering
cohesive soils with methylene blue value. Adv Civil Eng 2018;2018:7 p. https://doi.
practice. New Jersey: John Wiley & Sons; 2012.
org/10.1155/2018/9213674.
[14] Fredlund DG, Xing A. Equations for the soil-water characteristic curve. Can Geotech [50] Zhou AN, Sheng D, Carter JP. Modelling the effect of initial density on soil-water
J 1994;31:533–46.
characteristic curves. Géotechnique 2012;62(8):669–80.
[15] Fredlund MD, Wilson GW, Fredlund DG. Use of the grain-size distribution for es-
timation of the soil-water characteristic curve. Can Geotech J 2002;39:1103–17. Further reading
[16] Gallipoli D, Wheeler SJ, Karstunen M. Modelling the variation of degree of sa-
turation in a deformable unsaturated soil. Géotechnique 2003;53(1):105–12. [6] Barnston AG. Correspondence among the Correlation, RMSE, and Heidke Forecast
[17] Goh SG, Rahardjo H, Leong EC. Shear strength equations for unsaturated soil under Verification Measures; Refinement of the Heidke Score. Weather Forecasting
drying and wetting. Geotech Environ Eng 2010:136. 1992;7:699–709.
[18] Goh SG, Rahardjo H, Leong EC. Shear strength of unsaturated soils under multiple

11

You might also like