Publications: Water Resources Research
Publications: Water Resources Research
Publications: Water Resources Research
1. Introduction
Received 19 APR 2017
Accepted 18 AUG 2017 Understanding the hydraulic conductivity of a geomaterial, noted as K [m/s], is crucial for various
Accepted article online 6 SEP 2017 aspects of hydrogeology and geotechnical engineering, such as for modeling the groundwater flow. It
Published online 10 SEP 2017 is recognized that the hydraulic conductivity of a sediment sample is related to its pore structure, which
Corrected 22 JAN 2018
is difficult to measure. Pore structure is however intrinsically dependent on soil particle size distribution
(PSD), and estimating hydraulic conductivity from empirical equations based PSD is still widely adopted
This article was corrected on 22 JAN
in the nowadays [Song et al., 2009; Vienken and Dietrich, 2011; Lu et al., 2012; Riva et al., 2014]. The earli-
2018. See the end of the full text for
details.
est empirical equation of estimation hydraulic conductivity from particle size dates back to the work of
Hazen [1892], in which the hydraulic conductivity was expressed proportional to the squared grain size
at 10% passing (noted as d10 , unit in meter to obtain K in m/s). By considering unit consistency, it may
be written as:
g 2
K5CH d10 (1)
v
where g is the gravitational acceleration [m/s2] and m is the fluid kinematic viscosity [m2/s]
(m50:8931026 m2 =s at 25 C for water). CH is a unitless coefficient about 6:5431024 [Harleman et al., 1963].
Other typical equations to estimate hydraulic conductivity from grain size, but considering also the effects
of particle size uniformity and porosity have been also developed later. The equation of Beyer [1964], for
instance, is mostly known to account the effect of particle size uniformity:
g 500 2
K5CB log d10 (2)
v Cu
in which the coefficient CB is 631024 (unitless) and the coefficient of uniformity Cu (unitless) is the ratio
of grain size at 60% passing and grain size at 10% passing (Cu 5 dd60
10
). Furthermore, the Kozeny-Carman
C 2017. American Geophysical Union.
V model [Kozeny, 1927, 1953, Carman, 1937, 1956] is another classical model to embed the porosity
All Rights Reserved. effect:
g n3
K5CK d2 (3)
v ð12nÞ2 10
where the coefficient CK is 1=180 determined by flow in capillary tubes or beds of spheres and n is the
porosity.
However, these different models are developed based on a limited number of samples and may therefore
not be applied to all conditions. It should also be noted that the equations above are mostly empirical, and
the physical background is insufficiently considered. It is better to use physically independent variables in
the empirical equations (such as that soil porosity may be related to PSD and grain shape), and not only d10
but also the mean size of a PSD may also affect the hydraulic conductivity value. Rosas et al. [2014] tested
431 samples (0:05 mm < d10 < 0:83 mm, 0:09 mm < d60 < 4:29 mm, 1:3 < Cu < 18:3) from global sites
with different depositional environment based on the constant water head method. The results are normal-
ized by using the kinematic viscosity of water at 25 C (m50:8931026 m2 =s). These results can be regarded
as a database highly representative for the analysis of the relationship between PSD and hydraulic conduc-
tivity. For the purpose to consider the physical basis, the classic dimensional analysis (also known as Buck-
Q
ingham’s theorem, Buckingham [1914]), could be a useful tool to investigate an hypothetical physical
relationship. Indeed, this methodology reduces the number of parameters, the reduction being equal to the
rank of the so-called dimensional matrix (see next section). By using the dimensionless groups, it can sim-
plify the regression analysis of an experiment data set [Vignaux, 1992]. In this technical note, we will carry
out a dimensional analysis to explore the relationship between PSD and hydraulic conductivity and there-
fore clarify the controlling parameters. In particular, the effects of mean grain size, grain size uniformity,
and porosity will be integrated in the analysis. Then, a regression analysis will be conducted on the data-
base (including 431 samples) to obtain novel equations toward hydraulic conductivity estimation. The new
equation will then be benchmarked thanks to experimental results beyond the database to prove its
applicability.
Since P must have no dimension, these exponents are linked by dimensional equations (or dimensional
constraints): for each base dimension involved in the physical problem such as mass (M), length (L), time (T),
etc., a dimensional equation can be written in the following form:
X
dij sj 50 (5)
where dij is the exponent of dimension i in the parameter Xj . These dij coefficients define the so-called
dimensional matrix with p rows (p is the number of base dimensions, for instance, three in a mechanical
problem driven by mass, length, and time) and n columns (n is the number of physical parameters).
Let us now recall the results of the Buckingham’s theorem: the n dimensional parameters are linked by k
dimensional equations, where k is the rank of the dimensional matrix (so k p), which reduces the total
number of independent quantity terms to i5n2k. Moreover, the theorem also states that any physical
relationship f linking a dependent dimensional variable X0 and n dependent dimensional parameters such
as below:
X0 5f ðX1 ; X2 ; ; Xn Þ (6)
can be captured by another mathematical relationship F linking the i11 dimensionless terms as:
P0 5F ðP1 ; P2 ; ; Pi Þ (7)
In dimensional analysis, the variables on the right side of the above equation are assumed to be indepen-
dent from one another. However, it can be imaged that the porosity n of a granular material may relate to
PSD, especially to the grain size uniformity Cu . Indeed, pores between large grains will be filled with finer
particles if the grain size polydispersity is high. Accordingly, Vukovic and Soro [1992] found that the porosity
n is indeed a function of Cu empirically given by:
n5a 11bCu (9)
where a and b are unitless constants. This relationship is graphically shown in Figure 1 for the 431 unlithi-
fied sediment samples of Rosas data base [Rosas et al., 2014] and regressed to fit equation (9), which leads
to fitting parameters a50:2 and b50:93. The fitted line represents the porosity deduced from grain size uni-
formity, which can be referred as the intrinsic porosity. However, we can see that the measured porosity
has a variation around the intrinsic porosity; it will be considered in a latter part.
In the first step, it is assumed that the variation of the porosity around its intrinsic value do not affect the
hydraulic conductivity estimation. The number of independent parameters in equation (8) can therefore be
reduced from 8 to 7. The fundamental dimensions are length (L) and time (T) and the symbol – means
dimensionless. In this particular case, the number of degrees of freedom in equation (8) can be reduced
from 7 to 5 (i.e., the rank k is equal to 2). Additionally, as the coefficient of curvature Cc is not provided in
the database of Rosas et al. [2014] (d30 is not available), we do not consider here the effect of this parameter.
The remaining five variables may then be grouped in three-dimensional groups, written as follows thanks
to dimensional analysis:
K g
21 m 5F 23 2 ; Cu (10)
d60 d60 m
K
in which P0 5 d21 is a dimensionless hydraulic conductivity, P1 5 d23g m2 as a normalized gravity effect and P2
60 m 60
5Cu representing the grain size polydispersity effect (Cu is already dimensionless!). By assuming constant
temperature, m does not change for a given liquid. Since g is also a constant, d23g m2 actually represents the
60
effect of mean particle size (or mean particle volume as the exponent is in the third order).
To go one step further, let us remind that the set of dimensionless groups are linearly independent, which
means any dimensionless group can be replaced by a product of powers of the existing dimensionless
where C1 and a are fitting parameters. The best fitted a by the least square method is about 22.2. The
Hazen type equation can be recovered by taking a522 and the best coefficient for a522 is
C1 51:5231023 . Then, the hydraulic conductivity is:
g 2 g 2
K5C1 Cu a d60 C1 d10 (13)
m m
This also suggests that taking C1 51:5231023 instead of the Hazen value CH 56:5431024 may be more
appropriate for sediments in different depositional environments, according to the regression analysis on
the 431 samples.
One step further, the effect from d23g m2 now should be unraveled. In Figure 2b, the ratio between the mea-
60
sured hydraulic conductivity (Kmeasure ) from the Rosas’ data set (measured by a standard constant head per-
meameter following Wenzel [1942] and ASTM standard D2434–68, American Society for Testing and Materials
gd 3
[2006]) and the hydraulic conductivity predicted by equation (13) (Kpredict ) are depicted versus m260 for the
gd 3
431 sediments in a semilogarithm plot. It can be seen that for relatively large mean grain size (large m260 ),
equation (13) overestimates the hydraulic conductivity value and for finer materials equation (13) may have
3
gd60
Figure 2. Regression analysis on the two independent dimensionless variables. (a) Effect of Cu on K ; (b) Correction of m2 .
3
gd60
underestimation. The ratio of Kmeasure =Kpredict is inversely proportional to log10 m2 and may be expressed as
the following form:
3
21
Kmeasure gd60
5C2 log 10 2 (14)
Kpredict m
where C2 is estimated to be equal to 1.9 by fitting the database of 431 sediments. Then taking the correc-
tion from equation (14) on equation (13), K can be expressed as:
21 21
g 2 gd 3 g 2 gd 3
K5CW Cu a d60 log 10 260 CW d10 log 10 260 (15)
m m m m
where CW 5C1 C2 52:931023 and a 22. In the above equation, the effect of particle size polydispersity Cu
and the effect from the mean grain size d60 are already embedded with the effect of porosity regarded as a
dependent parameter on PSD. In reality, as can be seen in Figure 1, the porosity of a sediment sample may be
varied around the theoretical correlation due to the particle arrangement and stress state. Then a further cor-
rection may be considered by accounting the effect of the porosity variation. A parameter Dn, as the differ-
ence between the measured porosity and the intrinsic porosity in equation (9), is adopted to correct the
dimensionless hydraulic conductivity K . To keep the model simple Dn may be assumed to be a linear rela-
tionship with the K value. Therefore, another possible form of the hydraulic conductivity equation could be:
21 !
3
a gd60 g 2
K5 CW Cu log 10 2 1bDn d (16)
m m 60
where b is a fitting parameter. It should be noted that the linear relationship here may not be a rigorous
assumption and here it quantitatively reflects that Dn could be an additional parameter to develop. Due to
the limitation of the available data, we leave the Dn effect to future research.
10 -2 10 -2
(a) Predicted by Beyer model
(b)
Predicted by Hazen model
CB =6x10 -4
CH=6.54x10-4
Equality line
Regression line
10 -4 10 -4
Equality line
Regression line
slope=0.534
10 -5 10 -5
10 -5 10 -4 10 -3 10 -2 10 -5 10 -4 10 -3 10 -2
Predicted Value (m/s) Predicted Value (m/s)
10 -2 10 -2
(c) (d)
Predicted by Kozeny-Carman model Predicted by Eq. 13
CK =1/180 C1 =1.52x10-3 , a=-2
Measured Value (m/s)
10 -4 10 -4
10 -3 10 -3
10 -4 10 -4
Figure 3. Prediction of different models on the data set of Rosas et al. [2014]. (a) Hazen model; (b) Beyer model; (c) Kozeny-Carman model;
(d) Prediction of equation (13); (e) Prediction of equation (15); (f) Prediction of equation (16).
coefficient value will only shift the data set without changing the slope of regression line. This means that the
predictions of equations (15) and (16) are always better than other models (for this set of data).
As the data set of Rosas et al. [2014] contains hydraulic conductivity values for materials in various deposi-
tional environments and the samples are collected on global sites, the equations developed based on
dimensional analysis and this data set can be widely applicable to different sandy soils beyond this data set.
Validation of the new model can be carried out on other sandy soils in the literature (tested by the same
constant head method). Alyamani and Sen [1993] tested the hydraulic conductivity of 22 quaternary deposit
samples from Saudi Arabia. Wiebenga et al. [1970] tested 13 samples of quartz sands and silts from Australia.
Cabalar and Akbulut [2016] measured hydraulic conductivity values of 16 Narli sands and 16 crushed
stone sands from Turkey. Figure 4 compares the prediction values and the measured values of these three
data sets as a validation of the equation developed from dimensionless terms and regression analysis.
In Figure 4a, the prediction of Hazen equation is presented. The Hazen model has a good prediction for
10 0 10 0
Cabalar & Akbulut (2016) (a) Cabalar & Akbulut (2016) (b)
Alyamani & Sen (1993) Alyamani & Sen (1993)
-1 -1
10 Wiebenga et al. (1970) 10 Wiebenga et al. (1970)
Equality line Equality line
Regression line, slope=0.788 Regression line, slope=0.769
10 -3 10 -3
10 -4 10 -4
10 -3 10 -3
10 -4 10 -4
10 -5 10 -5
Predicted by Kozeny-Carman model Predicted by Eq. 13
-3
CK =1/180 C1=1.52x10 , a=-2
10 -6 10 -6
10 -6 10 -4 10 -2 10 0 10 -6 10 -4 10 -2 10 0
Predicted Value (m/s) Predicted Value (m/s)
10 0 10 0
Cabalar & Akbulut (2016) (e) Cabalar & Akbulut (2016) (f)
Alyamani & Sen (1993) Wiebenga et al. (1970)
-1 -1
10 Wiebenga et al. (1970) 10 Equality line
Equality line Regression line, slope=0.90
Regression line, slope=0.91
Measured Value (m/s)
10 -2 10 -2
10 -3 10 -3
10 -4 10 -4
Figure 4. Prediction on experiment results of Wiebenga et al. [1970], Alyamani and Sen [1993], and Cabalar and Akbulut [2016]. (a) Hazen model;
(b) Beyer model; (c) Kozeny-Carman model; (d) Prediction of equation (13); (e) Prediction of equation (15); (f) Prediction of equation (16).
these three data sets, and the slope of the regression line is slightly smaller than the equality line. The Beyer
model in Figure 4b shows that this model overestimates the hydraulic conductivity values and the slope of
the regression line is not improved. Estimation results of the Kozeny-Carman model is compared in Figure
4c. The porosity property is not available in the test results of Alyamani and Sen [1993] thus only the experi-
ments of Wiebenga et al. [1970] and Cabalar and Akbulut [2016] are compared. It can be seen that the
Kozeny-Carman model underestimates most of the results. The predictions by equations (13) and (15) are
displayed in Figures 4d and 4e, respectively. Equation (13) is a similar type of Hazen equation with a differ-
ent coefficient, and it also has a fair estimation on this two data sets. Furthermore, equation (15), in which
both of the effects of particle size uniformity and gravity (proportional to mean particle volume) are consid-
ered, has the best prediction among the models. This proves the good applicability of equation (15) in dif-
ferent deposit materials. Figure 4f presents the prediction of equation (16) in which the effect of porosity
variation around the so-called intrinsic porosity is considered. But the prediction accuracy is not obviously
improved from equation (15). This means in practice equation (15) could be the best choice.
4. Conclusions
We applied dimensional analysis in the relationship between PSD and hydraulic conductivity and used a
database containing 431 samples to carry out a regression analysis. The coefficient of uniformity Cu quanti-
gd 3
fying the grain size polydispersity and the dimensionless term of m260 are found to be the main controlling
parameters of the estimation. After regression analysis, new equations for estimating hydraulic conductivity
2
are developed. The classic Hazen type equation (K is propotional to d10 ) is found to be in consistence with a
simplified form of the regressed equation mainly considering Cu effect (equation (13)). By comparing with
the prediction performances of the three classic equations in the literature based on the database of Rosas,
gd 3
the new equation with the effect of m260 (equation (15)) improved the prediction significantly. Moreover, by
introducing the parameter of Dn, which is the difference between the measured porosity and the theoreti-
cal correlation with PSD, equation (16) further improved the prediction accuracy and reduced the variance
slightly. To validate the applicability of the new model beyond the database, the new equations are
employed to estimate hydraulic conductivity values of two set of experimental results in the literature. We
gd 3
find that equation (15), with the term m260 embedded, which is representing the dimensionless gravity effect
(proportional to mean grain volume), has the best estimation accuracy. This implies the wide application
prospect of the new equations proposed in this report.
Acknowledgment References
This work was financially funded by a
grant from F.R.S-FNRS of Belgium Alyamani, M. S., and Z. Sen (1993), Determination of hydraulic conductivity from complete grain-size distribution curves, Ground Water,
(PDR.T.1002.14). All data analyzed in 31(4), 551–555, doi:10.1111/j.1745-6584.1993.tb00587.x.
this paper can be found in the cited American Society for Testing and Materials (2006), Standard Test Method for Permeability of Granular Soils, ASTM Standard D2434–68, Am.
references. Soc. for Test. and Mater., West Conshohocken, Pa.
Beyer, W. (1964), Zur Bestimmung der Wasserdurchlassigkeit von Kieson und Sanduen aus der Kornverteilung, Wasserwirt. Wassertech., 14, 165–169.
Buckingham, E. (1914), On physically similar systems; illustrations of the use of dimensional equations, Phys. Rev., 4(4), 345–376, doi:
10.1103/PhysRev.4.345.
Cabalar, A. F., and N. Akbulut (2016), Evaluation of actual and estimated hydraulic conductivity of sands with different gradation and
shape, Springerplus, 5(1), 820, doi:10.1186/s40064-016-2472-2.
Carman, P. (1937), Fluid flow through granular beds, Trans. Chem. Eng., 15, 150–166.
Carman, P. C. (1956), Flow of Gases Through Porous Media, Butterworth Sci. Publ., London.
Cheng, Y.-T., and C.-M. Cheng (2004), Scaling, dimensional analysis, and indentation measurements, Mater. Sci. Eng. R, 44(4–5), 91–149, doi:
10.1016/j.mser.2004.05.001.
Harleman, D., P. Mehlhorn, and R. Rumer (1963), Dispersion-permeability correlation in porous media, J. Hydraul. Div. Am. Soc. Civ. Eng.,
89(2), 67–85.
Hazen, A. (1892), Physical Properties of Sands and Gravels With Reference to Their Use Infiltration, Massachusetts State Board of Health, Bos-
ton, Mass.
€
Kozeny, J. (1927), Uber kapillare leitung des wassers im boden: (aufstieg, versickerung und anwendung auf die bew€asserung), H€ older-Pich-
ler-Tempsky, 136, 271–306.
Kozeny, J. (1953), Das Wasser im Boden. Grundwasserbewegung, in Hydraulik, edited by J. Kozeny, pp. 380–445, Springer, Vienna.
Lu, C., X. Chen, C. Cheng, G. Ou, and L. Shu (2012), Horizontal hydraulic conductivity of shallow streambed sediments and comparison with
the grain-size analysis results, Hydrol. Processes, 26(3), 454–466, doi:10.1002/hyp.8143.
Riva, M., X. Sanchez-Vila, and A. Guadagnini (2014), Estimation of spatial covariance of log conductivity from particle size data, Water
Resour. Res., 50, 5298–5308, doi:10.1002/2014WR015566.
Rosas, J., O. Lopez, T. M. Missimer, K. M. Coulibaly, A. H. A. Dehwah, K. Sesler, L. R. Lujan, and D. Mantilla (2014), Determination of hydraulic
conductivity from grain-size distribution for different depositional environments, Ground Water, 52(3), 399–413, doi:10.1111/gwat.12078.
Song, J., X. Chen, C. Cheng, D. Wang, S. Lackey, and Z. Xu (2009), Feasibility of grain-size analysis methods for determination of vertical
hydraulic conductivity of streambeds, J. Hydrol., 375(3–4), 428–437, doi:10.1016/j.jhydrol.2009.06.043.
Vienken, T., and P. Dietrich (2011), Field evaluation of methods for determining hydraulic conductivity from grain size data, J. Hydrol.,
400(1–2), 58–71, doi:10.1016/j.jhydrol.2011.01.022.
Vignaux, G. (1992), Dimensional analysis in data modelling, in Maximum Entropy and Bayesian Methods, edited by C. R. Smith, G. J. Erickson
and P. O. Neudorfer, pp. 121–126, Springer, Netherlands.
Vukovic, M., and A. Soro (1992), Determination of Hydraulic Conductivity of Porous Media From Grain-Size Composition, Water Resour. Publ.,
Littleton, Colo.
Wang, J.-P., N. Hu, B. François, and P. Lambert (2017), Estimating water retention curves and strength properties of unsaturated sandy soils
from basic soil gradation parameters, Water Resour. Res., 53, 6069–6088, doi:10.1002/2017WR020411.
Wenzel, L. (1942), Methods for determining permeability of water-bearing materials, with special reference to discharging-well methods,
with a section on direct laboratory, U.S. Geol. Surv. Water- Supply Pap. 887.
Wiebenga, W. A., W. R. Ellis, and L. Kevi (1970), Empirical relations in properties of unconsolidated quartz sands and silts pertaining to water
flow, Water Resour. Res., 6(4), 1154–1161, doi:10.1029/WR006i004p01154.
Erratum
In the originally-published version of this article, the denominator of equation (3) was incorrectly given as
(1–n)3. This error has been corrected, and this version may be considered the official version of record.
Article: 22879
Dear Author,
Please refer to the query reference callout numbers in the page proofs and respond to each by
marking the necessary comments using the PDF annotation tools.
Please remember illegible or unclear comments and corrections may delay publication.
AQ2 AUTHOR: Table 1 is cited in the text but not provided. Kindly check and
provide the table.
AQ3 AUTHOR: Please provide publisher name and publisher location for
Hazen [1892].
AQ5 AUTHOR: Please provide sponsor name and sponsor location for Wenzel
[1942].
AQ6 AUTHOR: Please confirm that given names (red) and surnames/family
names (green) have been identified correctly.
AQ7 AUTHOR: Please complete the Publication Fee Consent Form included
with your article and return to the Production Editor with your proofs.
Please use any version of Adobe Acrobat Pro or Reader 11 or higher to
edit the form. Reader can be downloaded free at: http://www.adobe.com/
products/acrobat/readstep2.html.