You are on page 1of 14

Journal of Hydrology 272 (2003) 36–49

www.elsevier.com/locate/jhydrol

Linking hydraulic conductivity and tortuosity parameters to pore


space geometry and pore-size distribution
R.W. Vervoort*, S.R. Cattle
School of Land, Water and Crop Sciences, Bldg. A03, The University of Sydney, 2006 NSW, Sydney, Australia

Abstract
To enable a link between the pore size distribution of the soil measured in two dimensions and the hydraulic conductivity, a
factor, which includes the connectivity and tortuosity of the pore network in three dimensions, is needed. In order to increase
confidence in the models which predict hydraulic conductivity, this paper aims to gain insight into relationship between
tortuosity and connectivity parameters used in statistical models of hydraulic conductivity and soil structural form
measurements derived with image analysis. The analysis was based on a dataset of the hydraulic and soil structural properties of
19 Vertisol samples from Eastern Australia. The soil structural data were measured on large (200 mm height and 150 mm
diameter) cores at 5 – 7 depth slices. Several soil structural form parameters, including a pore size distribution, were calculated
using an image analysis program (Solicon v2.1 ). The hydraulic conductivity model developed by Kosugi [Soil Sci. Soc. Am. J.
63 (1999) 270] based on the lognormal pore size distribution was fitted to measured hydraulic conductivities. Hydraulic
conductivity was strongly related to porosity and a measure of connectivity in the horizontal direction (pore genus). The
tortuosity parameters a and b from the Kosugi [Soil Sci. Soc. Am. J. 63 (1999) 270] model were related to the mean pore size,
with increasing mean pore size indicating decreasing tortuosity of the sample.
q 2002 Elsevier Science B.V. All rights reserved.
Keywords: Pore size distribution; Hydraulic conductivity; Tortuosity; Connectivity

1. Introduction matching parameters often defined as connectivity and


tortuosity. Without these factors the pore size distri-
1.1. Tortuosity and connectivity bution models would mimic the permeability of a
bundle of straight capillary tubes. Reality in a soil is
The hydraulic conductivity or permeability of the much more complex, with twisted and crooked pores,
soil is determined by the arrangement of the pores and dead-ending or connecting to other pores. This means
solids, the pore and solid space geometry. Statistical there is a need to scale the permeability from the
models for the permeability and hydraulic conductivity capillary tube model to include increased path length
of soil are based on the distribution of pore sizes (e.g. due to crookedness of the path (tortuosity) or lack of
Mualem, 1976; Kosugi, 1999), but the models include connection between points in the soil (connectivity). In
theory, the two parameters should be inversely related,
with highly connected (or highly porous) soils having
* Corresponding author. Tel.: þ61-2-9351-8744; fax: þ 61-2-
9351-5108.
low tortuosities and vice versa (Vogel, 1997). For the
E-mail address: w.vervoort@acss.usyd.edu.au (R.W. purpose of this paper we will consider the two
Vervoort). parameters interchangeable.
0022-1694/03/$ - see front matter q 2002 Elsevier Science B.V. All rights reserved.
PII: S 0 0 2 2 - 1 6 9 4 ( 0 2 ) 0 0 2 5 3 - 6
R.W. Vervoort, S.R. Cattle / Journal of Hydrology 272 (2003) 36–49 37

The connectivity and tortuosity parameters in relative saturation (Se):


hydraulic conductivity models are generally rather ðSe  ð1 2
obscure matching parameters, which have been Kr ¼ Se l 21
h dSe 21
h dSe ð4Þ
defined differently throughout the literature. The 0 0
origin lies in the Kozeny – Carman equation (Bear,
in which t ¼ Sel or t ¼ u l
1972):
In these equations l is an empirical parameter,
c0 tf3 which is different for each soil, and generally we
k¼ ð1Þ consider t , 1 (Mualem and Dagan, 1978). Higher
s2
values of l are thus related to lower values of t and
This equation relates the permeability of the soil to the thus to higher effective path lengths (following
average hydraulic radius of the pore space. Eq. (2)). This concept of tortuosity is based on the
The hydraulic radius is defined by the porosity (f ) fact that pores are irregular and upon drying, water
and the average pore perimeter (s ). The equation also will concentrate in small angles and crevices of the
includes two matching factors: tortuosity (t ) and a pore system as water films (e.g. Tuller et al., 1999).
pore shape factor (c0). These factors reduce the The result is an increased path length, less connection
‘straight and circular shaped pore’ value of the and a more tortuous path with decreasing water
permeability to a value supposedly representative content (e.g. Vogel, 1997). This definition of
for the actual pore space geometry. Tortuosity in this tortuosity is the macroscopic approach (Mualem and
equation is also defined in terms of the ‘effective path Dagan, 1978), since it reduces the hydraulic
length’ (Le) Bear (1972): conductivity at the ‘whole soil’ level. It also seems
 2 a description of pore roughness, which is the
L
t¼ ,1 ð2Þ definition used in the applications of fractal geometry
Le to the soil water characteristic and hydraulic conduc-
where L is the length of the sample. The squaring tivity (Crawford, 1994; Giminez et al., 1997), with
evolves from the fact that the path length affects both pore with higher pore roughness having a higher
the driving force (potential gradient) at the local scale tortuosity and lower connectivity.
and the velocity as defined by Poiseuille’s law (e.g. Comparing the different statistical models, Mualem
Bear, 1972, p. 110). Another common definition of and Dagan (1978) point out that another (microscopic)
tortuosity is the ratio of the effective path length to the representation of tortuosity can be based on Fatt and
sample length (Kutilek and Nielsen, 1994): Dijkstra (1951). In this case the tortuosity is inversely
proportional to a power function of the pore radius (r ),
Le which leads to the definition of the effective pore size
t¼ .1 ð3Þ
L (re):
This means tortuosity would be in the denominator of re2 ¼ r 2þb and t ¼ rb ð5Þ
Eq. (1) since t . 1.
The Kozeny – Carman model is limited to the where b is again an empirical parameter. Theoretically
prediction of the permeability of a sample at this means that the tortuosity is a function of the pore
saturation. Later models, based on the work by Childs size, but is constant for each soil. Considering the
and Collis-George (1950), Millington and Quirk suggested power relationship, this definition seems
(1960) and Marshall (1958), expand the concepts to again fractal (Crawford, 1994), but at the pore, rather
include unsaturated hydraulic conductivities or per- than the whole soil level. Note that for both Eqs. (4) and
meabilities at lower water contents. In these models (5) the tortuosity is generally regarded to be smaller
tortuosity is defined in terms of an effective area than unity, that is, it reduces the hydraulic conductivity
available for flow through a sample section, and as calculated with the capillary tube model to its ‘real’
such affecting the relative hydraulic conductivity. value. This would be consistent with the formulation of
Mualem (1976) suggests the following model for the tortuosity in Eq. (2), but not with the definition in
relative hydraulic conductivity (Kr) as a function of Eq. (3). The tortuosity or connectivity should also be
38 R.W. Vervoort, S.R. Cattle / Journal of Hydrology 272 (2003) 36–49

decreasing with increasing saturation or pore size the pore size. In theory, high porosity media tend to be
(Vogel, 1997). This means we can define the following more connected and should have a lower tortuosity,
theoretical ranges for the parameters: and thus the percolation threshold would define the
maximum tortuosity within the sample. Above the
l,0 and 0 , Se , 1
percolation threshold the tortuosity should decrease to
b.0 if r , 1 a minimum value. This is a similar macroscopic
definition of tortuosity such as Eq. (4). Network
b,0 if r . 1
models predicting flow and permeability based on this
However, it has been indicated before that many of the theory have been explored in soil physics and
fitting results on measured hydraulic conductivity hydrology (Berkowitz and Balberg, 1993; Yanuka
values can indicate values for these parameters well et al., 1986; Vogel and Roth, 1998; Peat et al., 2000),
outside these theoretical ranges (Kosugi, 1999). but have generally remained a research tool due to the
Kosugi (1999) reformulated the general Mualem computational complexity. Future developments in
model for the relative hydraulic conductivity (Mua- algorithms and computational simplicity could see
lem and Dagan, 1978) to include the pore size more widespread use of these approaches.
distribution: Both the current macroscopic and microscopic
2 ðr 3g approach for tortuosity relate to the magnitude of the
b
6 0 r gðrÞdr 7 relative hydraulic conductivity. This means that the
Kr ¼ Sea 64 ð1
7
5 ð6Þ parameters represent relative tortuosities. However,
gðrÞdr we would still expect a soil with a more connected
0
pore space to have lower relative tortuosities, since for
Here, g(r ) is the pore size distribution and Se is the each pore size class there will be more connections
relative saturation. The parameter a is equivalent to l and thus a higher relative tortuosity. Even though
in Eq. (4), and is thus the equivalent macroscopic most of the theory behind the hydraulic conductivity
description of tortuosity, similar to the adjustment in model suggests a clear relationship between soil
the Kozeny –Carman equation, while b is related to b structure and tortuosity, in most cases the parameters
in Eq. (5) and is the equivalent microscopic descrip- in models are treated as fitting parameters, and no
tion of tortuosity. The parameter g is related to the relation with soil physical or pore space geometry is
correlation between pores on either side of a slab soil suggested. However, ensuring that these parameters
and thus again to the effective pore radius available have physical significance will improve our ability to
for flow (Mualem and Dagan, 1978). According to scale hydraulic conductivities across spatial and
Kosugi (1999), the model is rather insensitive to this temporal scales.
parameter and it can be set to 1. We have followed this
approach in this paper. 1.2. Image analysis of pore and solid space geometry
Fatt (1956) has suggested a different model based
on a network of interconnected pores based on the In order to identify the pore size distribution and
principles of percolation theory (Broadbent and such parameters as connectivity and tortuosity, image
Hammersley, 1957). In this approach randomness is analysis of soil sections has been employed (Vogel,
assigned to the medium, and not the fluid. The 1997; Moreau et al., 1999). Berryman and Blair (1987)
probability of open and closed bonds is defined and calculated pore perimeter and pore area from thin
this defines the overall probability of a connected sections to calculate permeabilities using the K – C
network of pores through a medium. No flow in the model, but no attempt was made to calculate
medium would occur unless a connected network connectivity or tortuosity. Fractal approaches have
exists. Golden (1980) laid the theoretical groundwork also leaned on image analysis to calculate fractal
for the use of network percolation theory in dimensions of pore space and solid (e.g. Anderson
unsaturated media, by describing the similarity et al., 1996; Crawford, 1994). Consecutive thin
between open and closed bonds by basing the sections of a soil block can be used to identify the
probability of open bonds on the suction, and thus to connectivity as defined by the Euler – Poincaré
R.W. Vervoort, S.R. Cattle / Journal of Hydrology 272 (2003) 36–49 39

characteristic, which is a measure of the number of a rapid and useful technique to actually measure
independent loops in a structure (e.g. Vogel, 1997; tortuosity. Considering the complexity of the pore
Moreau et al., 1999). Vogel (1997), using this method, geometry and the anisotropy in the vertical direction,
demonstrated that the connectivity indeed tended to this is not easy (Moreau et al., 1999; Perret et al.,
increase with higher saturations. However, the thin 1999; Vogel, 1997). The aims of this paper are
section method is labour intensive and not well suited twofold: the first is to describe the general soil
to large soil volumes. X-ray tomography can be used structural form of 19 Vertisol samples using image
on larger soil volumes (Perret et al., 1999), but requires analysis and to relate these measures to saturated
specialist equipment and costs and access limitations hydraulic conductivity; the second is to relate
can be significant. Both methods have delivered insight measurements of soil structural form to tortuosity
in the magnitudes of parameters describing connec- and connectivity parameters in the general model for
tivity and tortuosity. hydraulic conductivity (e.g. Eq. (6)). This information
Pore size distributions can be derived from a slice might not be of much use in terms of predicting
through the medium, but this would assume an hydraulic conductivity (in contrast to pedotransfer
isotropy in the vertical direction, which is generally functions based on soil properties, e.g. Minasny and
not true. Stereology can be used to infer three- McBratney (2000)), since it is more difficult to
dimensional properties from two-dimensional pic- perform the image analysis than it is to measure
tures (Serra, 1982), but the results are statistical hydraulic conductivity. However, it might increase
estimations of the third dimension based on the confidence in the physical model, understanding of
measurement in two dimensions. Averaging the pore structural form effects on its behaviour, and its
size distributions of different slices might improve the applicability to different situations. Alternatively it
estimation of the real pore size distribution, but will allow interpretation of the parameters in the
increases the effort and cost. It can also be argued that hydraulic conductivity model in terms of soil
the connectivity and tortuosity parameters in the structural form.
hydraulic conductivity model will correct for the lack
of representativeness of the pore size distribution.
Although most work has concentrated on defining the
2. Materials and methods
pore space geometry, it is very well possible that the
surrounding solid space defines the pore space and as
such the permeability, tortuosity and connectivity. 2.1. Sites and sampling

1.3. Aims of this paper Fourteen sites, located in five different cotton-
growing areas in New South Wales and Southern
Trying to predict the hydraulic conductivity from Queensland (Australia), have been sampled under this
pore or solid space geometry, we are faced with two project. All soils are classified as Vertisols (Soil
problems. The first one is the problem caused by the Survey Staff, 1996) or Vertosols (Isbell, 1996),
anisotropy of the medium in the horizontal direction. indicating soils with more than 35% clay in the B-
This means a ‘representative’ pore size distribution horizon and the existence of cracks, lenticular peds or
needs to be sampled from a representative elementary slickensides. Two of the sites consisted of soil pits
volume (REV, Bear, 1972). This problem has down to 2 m, in which different layers were sampled,
received ample attention in the literature. The other adding five subsoil samples leading to a total of 19
is the lack of information on the pore or solid sampling points. The range of clay contents in the
geometry in the vertical direction once the pore size samples was 39 –73%, with a mean value of 57%. The
distribution is known, e.g. the connectivity and sites were all located on irrigated cotton properties
tortuosity. This problem is much less well studied. and, within each field, were located at a minimum of
Two approaches are generally taken; the first one is to 100 m from the tail drain or head ditch. At each site a
treat these values as a fitting parameter (the most row in the cotton field was selected to perform the
common approach); the second approach is to develop experiments and sampling. Plant rows were selected
40 R.W. Vervoort, S.R. Cattle / Journal of Hydrology 272 (2003) 36–49

for the main measurements to avoid complication due with a slow-curing resin mixture containing an
to slaking and wheel traffic in furrows. ultraviolet (UV)-fluorescent yellow dye. The fluid
components of this mixture, as detailed in McBratney
2.2. Hydraulic conductivities et al. (1992), are resin, diluent, hardener and opacifier,
mixed in the ratios of 34:34:32:1. Considering the
On the row, infiltration was characterised using shrink-swell nature of Vertisols, drying of the cores
disc permeameter measurements at five different between excavation from the field and application of
supply potentials (2 150, 2 100, 2 70, 2 50 and the resin could have influenced the pore geometry.
2 20 mm). The infiltration data were used to calculate Soil wetness of the cores was therefore measured by
hydraulic conductivities at the mid points between the sampling some soil from the top and bottom of the
supply potentials (2 35, 2 60, 2 85 and 2 125 mm) core just before the application of the resin, and
and at the 2 20 and 2 150 mm supply potential subsequently oven dried at 105 8C. These values were
(Ankeny et al., 1991). To calculate the saturated compared to the wetness of samples taken in the field
hydraulic conductivity (Ks), Gardner’s equation for at the time of sampling. Following the hardening of
hydraulic conductivity (Gardner, 1958) was fitted to the applied resin mixture, each soil core was ground
the data using standard least squares techniques and down in horizontal section using an angle grinder, and
extrapolated towards a zero supply potential. This images of the exposed surface captured with a digital
extrapolation of the unsaturated hydraulic camera under UV light at 10, 20, 30, 40, 60, 80 and
conductivities is probably representative for the soil 100 mm depth. In some cases only the 20, 40, 60, 80
matrix and might underestimate the true Ks. Table 1 and 100 mm depth were collected and on some
gives an overview of the hydraulic conductivity data samples every 10 mm between 0 and 100 mm was
used in this study. collected. The colour images (composed of yellow
pores and black solids) captured from each core were
2.3. Soil structural analysis downloaded to a computer and segmented into black
and white binary images. These images where then
At each site, two large, undisturbed soil columns analysed using the program Solicon v2.1 (Cattle et al.,
(150 mm diameter and 200 mm height) were exca- 2001) whose predecessor Structura and the under-
vated from the same row where the hydraulic lying stereological model have been described earlier
conductivity measurements were carried out. The (McBratney et al., 1992; Anderson et al., 1996). Using
soil cores were encased in polyvinylchloride (PVC) various pixel-counting procedures, Solicon v2.1 was
piping and transported to the laboratory. In the used to estimate the structural parameters of porosity,
laboratory, the cores were irrigated to saturation surface area (SA), mean pore intercept length (MPIL),
mean solid intercept length (MSIL), pore star length
Table 1 (PSL), solid star length (SSL), pore genus (PG) and
An overview of the hydraulic conductivity dataset used in this study solid genus (SG). In addition, distributions of the pore
sizes in each layer were made using the histogram
Supply Geometric Max Min Geometric
potential mean (mm h21) (mm h21) SD function for the pore intercept length (PIL). Most of
(mm) (mm h21) the images were between 650 £ 650 and 750 £
750 pixels for a 100 mm £ 100 mm image. This
Fitted Ks 107.2 1224.0 1.0 0.72 meant that one pixel represented a pore diameter
220 25.6 189.0 0.59 0.58 between 0.15 and 0.13 mm. Hence, the porosity
235 11.2 60.0 0.37 0.53
calculated in this manner is probably representative
260 3.4 31.8 0.14 0.59
285 0.87 5.51 0.10 0.44 for the macroporosity (Beven and Germann, 1982;
2125 0.27 1.99 0.33 0.51 Watson and Luxmoore, 1986). Porosity is calculated
2150 0.22 1.66 0.03 0.48 as the number of pixels representing pore space
The table lists geometric means, maxima, minima and divided by the total number of pixels in the image,
geometric standard deviations (SD) of the hydraulic conductivity while SA, a measure of the pore/solid interface, is
(K ) at saturation (Ks) and at the six other supply potentials. calculated from the length of the boundary between
R.W. Vervoort, S.R. Cattle / Journal of Hydrology 272 (2003) 36–49 41

pixels representing pore space and pixels representing unsaturated hydraulic conductivities (K(h) ). We used
solids. The MPIL is a measure of the one-dimensional   
1 bs
continuity of pores, and is estimated by counting and KðhÞ ¼ Ks Sea erfc erfc21 ð2SeÞ þ pffiffi
2 2
averaging the continuous lengths of pixels represent- ð8Þ
 
ing pore space in four directions (0, 45, 90 and 1358), 1 lnðh=hm Þ
over the entire image. In terms of pore sizes, the MPIL Se ¼ erfc pffiffi
2 s 2
represents a mean pore diameter. The MSIL is
calculated in the same fashion, but for pixels in which a and b are fitting parameters, h is the
representing solids. PSL and SSL are estimates of potential at which the relative conductivity is
the average size of each phase, and are calculated as calculated, hm is the potential related to m, Ks is the
the expected continuous length of pore or solid saturated hydraulic conductivity and Se is the relative
encountered in any direction from a random point saturation. Nineteen pairs of a and b were thus
within that phase (Serra, 1982). The PG and SG are calculated, one pair for each of the PIL distributions
defined as a measure of the number of independent calculated with Solicon v2.1. These pairs of par-
loops (Vogel and Roth, 1998). Solicon v2.1 uses the ameters would then represent the tortuosity and
algorithm developed by Chen and Yan (1988) to connectivity of unobserved section of the samples
calculate the Euler– Poincaré number and subtracts below 20 mm.
the number of features in the image. This leads to a We also calculated the tortuosity as a function of
potential by comparing the measured hydraulic
value representing the number of loops per mm2.
conductivities with hydraulic conductivities calcu-
Further description of the stereological models used in
lated using a straight capillary pore model (Jury et al.,
Solicon v2.1 is made in McBratney and Moran (1990)
1991)
and McBratney et al. (1992).
As the resin was applied to the soil surface in each Kc ðhÞ
tðhÞ ¼ ð9Þ
core, the analysed pore phase features are representa- Km ðhÞ
tive of the connected macroporosity, or the part of the
where Km(h ) is the measured hydraulic conductivity,
macroporosity domain with a direct connection to the
and Kc(h ) is the conductivity calculated using (Jury
soil surface. All structural parameters were lognormal
et al., 1991):
distributed and the data were therefore log trans-
formed before taking averages. f2 Du X 1
Kc ðhÞ ¼
mrw g h2
2.4. Calculation of tortuosity where Du are the increments in water content, again
derived from the PIL distributions of the upper 20 mm
The log mean (m ) and standard deviation (s ) of of the samples, rw is the density of water, g the
the pore size distributions of the top 20 mm each core acceleration due to gravity, w the liquid – solid contact
were calculated by fitting a lognormal distribution angle and m the viscosity. This torturosity parameter
function to the cumulative PIL distribution calculated (based on the conductivity ratio) would again
using Solicon v2.1. We chose the top 20 mm of the represent the tortuosity and connectivity of the
core because this would be the pore size distribution unobserved section of the sample below 20 mm.
observed directly below the infiltration measurement,
which would be representative for most field
observations. Until recently (e.g. Perret et al., 1999), 3. Results and discussion
it was probably representative for most available
image analysis observations of pore size distributions, 3.1. Structural form attributes of Vertisols and
that is, from a two-dimensional slice through the relation to hydraulic conductivity
medium. Using this information, the statistical
hydraulic conductivity model developed by Kosugi The structural attributes of the sampled Vertisol
(1999) was subsequently fitted to the measured soil cores varied with depth from the soil surface
42 R.W. Vervoort, S.R. Cattle / Journal of Hydrology 272 (2003) 36–49

(Table 2). The so-called crumb layer, or self-mulched PG values generally decreased with depth (Table 2),
layer, is a distinctive feature of these soil types, and indicating less connectivity lower in the profile. Note
was typically 20 mm deep. This crumb layer is that it is probably a stronger indication of the decrease
characterised morphologically by abundant, small, in horizontal connectivity than of the actual vertical
irregularly shaped macropores interspersing a matrix connectivity in the core. However, if each value is
of medium granular (5 –10 mm diameter) and med- representative for the observed depth slice, than the
ium subangular blocky (10 – 20 mm diameter) aggre- average value over all slices for each core could be
gates. Such morphology was reflected in the image interpreted as the mean vertical connectivity. The
analysis data of all sites by large values of porosity relationship (Table 3) between porosity and PG
(mean, 0.23 mm 3 mm 23) and SA (mean, confirms that more porous samples tend to be more
, 0.92 mm2 mm23), and small values of MSIL connected. In addition, PG was well correlated to
(, 7 mm). In contrast the soil below 40 mm is much MSIL (Table 3), with the relationship indicating a
more dense which is characterised by smaller values decrease in horizontal connectivity with an increase in
of porosity, SA and larger values of MSIL. Due to the ped size. This relationship is not surprising since
irregular actual depth of the crumb layer, the 30 and MSIL and porosity were also strongly correlated. In
40 mm depths act as transition layers. Porosity and SA contrast, there was less of a relationship with pore-
decreased gradually with depth, while MSIL and SSL related parameters such MPIL and PSL, even though
gradually increased. In contrast, the pore related the PG itself is a pore-related parameter. The number
parameters MPIL and PSL appeared to decrease of independent loops in the pore domain is thus more
rapidly below the 10 mm depth, but attain an dictated by the space filling characteristics of the solid
approximate constant value below this layer. This phase, e.g. the average pedsize and porosity, than by
means that although the average ped size increases the average size of the pores. Solid genus also had a
only slowly, the packing density of the peds increases much lower correlation to the pore related structural
more sharply below the crumb layer. parameters, but was correlated to SA indicating that
Because there was some time between the field soils with a larger SA (more irregular shapes) tended
sampling and the impregnation of the cores, differ- to have more independent loops in the solid phase.
ences in soil water content could have caused changes The saturated hydraulic conductivity (Ks) was well
in structural form in these shrink swell soils. correlated to a number of the soil structural form
However, comparison of soil water contents measured parameters (Table 3). In particular, porosity and SA
in the field and before impregnation revealed only were positively correlated with Ks, with greater SA
slight differences with the mean field water content probably indicating more distinct pores. The mean
being 0.28 ^ 0.01 and being 0.27 ^ 0.01 g g21 just ped size (MSIL) was negatively correlated with Ks,
before impregnation. indicating a decrease in conductivity with an increase
PG describes the number of independent loops in in mean ped size. Note that most of the pore related
an image (Vogel, 1997). In the samples in this study, parameters (such as MPIL) had little or no
Table 2
Geometric means of image analysis parameters by depth

Depth Porosity SA MPIL MSIL PSL SSL PG SG


(mm) (mm3 mm23) (mm2 mm23) (mm) (mm) (mm) (mm) ( £ 1022 mm22) ( £ 1022 mm22)

10 0.17a 0.75a 0.98a 3.87a 2.37 12.03a 1.68a 12.24


20 0.15a 0.71ab 0.88ab 3.93a 1.86 12.46a 1.66a 12.03
30 0.06ab 0.36abc 0.75ab 7.78ab 1.46 21.93ab 0.51ab 8.11
40 0.07ab 0.37abc 0.77ab 7.31ab 1.52 22.26ab 0.67ab 8.06
60 0.06b 0.34abc 0.70b 8.91b 1.49 25.41b 0.45ab 9.30
80 0.05b 0.31bc 0.72ab 9.95b 1.56 28.52b 0.37b 9.08
100 0.03b 0.21c 0.71ab 13.98b 1.48 35.60b 0.22b 6.16

Different letters (a,b) in a column indicate difference in means at the p , 0.05 level using the Tukey–Kramer means separation test.
R.W. Vervoort, S.R. Cattle / Journal of Hydrology 272 (2003) 36–49 43

Table 3
Correlations between saturated hydraulic conductivity (Ks) and structural form parameters derived using Solicon v2.1

Log10 ðKs Þ Ks Porosity SA MPIL MSIL PSL SSL PG

Log10 ðKs Þ 1.00 – – – – – – – –


Ks 0.84 1.00 – – – – – – –
Porosity 0.72 0.62 1.00 – – – – – –
SA 0.73 0.57 0.90 1.00 – – – – –
MPIL 0.54 0.53 0.75 0.45 1.00 – – – –
MSIL 20.75 0.54 20.74 0.87 20.26 1.00 – – –
PSL 0.56 0.55 0.74 0.44 0.98 20.29 1.00 – –
SSL 20.68 0.50 20.78 0.91 20.26 0.96 0.27 1.00 –
PG 0.58 0.50 0.93 0.71 0.83 20.57 0.82 0.60 1.00
SG 0.50 0.33 0.44 0.76 20.13 20.76 0.08 0.78 0.18

relationship to Ks. After log transformation most of 3.2. Pore size distributions and hydraulic conductivity
the relationships were stronger, with the strongest fitting results
being the relationships between either log(MSIL) or
log(porosity) and log(K s) (r 2 ¼ 0:60 and 0.66, The PIL distributions were generally lognormally
respectively, Fig. 1). These relationships appear to distributed, which meant the lognormal distribution
indicate that, again, the size and the packing of the model fitted the data well, with an average root mean
solid phase (represented by MSIL and porosity), squared error (RMSE) of fit of 0.03. Using the
determining the connectedness of the pore space (PG), standard model suggested by Kosugi (1999), with
have the largest influence on the magnitude of Ks. parameter a set at 2 0.3 and b set at 1.8, hydraulic
This is further confirmed by the multi-linear relation- conductivities were predicted using the average pore
ship between the combination of porosity and SSL size distributions of the top 20 mm of the cores.
(being another measure of mean ped size) and The model was able to predict the hydraulic
Ks (logðKs ) ¼ log(porosity) þ log(SSL), r 2 ¼ 0:76). conductivities fairly well with an r 2 of 0.58 for

Fig. 1. Strong relationships between soil structural form and the saturated hydraulic conductivity are between Log10 ðKs Þ and Log(Porosity) and
Log(MISL).
44 R.W. Vervoort, S.R. Cattle / Journal of Hydrology 272 (2003) 36–49

Fig. 2. Predicted versus measured hydraulic conductivity (a) using the standard Kosugi model with a ¼ 20:3 and b ¼ 1:8 and (b) using the
Kosugi model, but fitting a and b for the individual samples.

the log – log relationship (Fig. 2a). Using the model in and b values are well within the range reported by
this way assumes a constant tortuosity and connec- Kosugi (1999). Using these values for a and b
tivity for all the samples. The general relationship thus improves the prediction of the hydraulic conductivity
reflects the relationship between the average pore considerably with and r 2 ¼ 0:91 for the log – log
distribution and the hydraulic conductivity, while the relationship (Fig. 2b), indicating that the spread in the
spread in the data is a reflection of the differences in data in Fig. 2a is indeed a reflection of the differences
tortuosity and connectivity of the individual cores in tortuosity and connectivity.
below 20 mm depth. There was little variation in the The question remains how, and whether, these
mean pore size (m ) and standard deviation of the pore parameters relate to measures of structural form (e.g.
size distribution (s ) of the cores calculated using a Table 2). The best relationship between measures of
lognormal pore size distribution model (Table 4), but structural form and tortuosity parameters was between
there was considerable variation in the a and b. All a MPIL and a (Table 5 and Fig. 3). At first glance the
relationship seems counter intuitive, smaller a should
Table 4 mean lower tortuosities and this would be expected in
Summary of tortuosity parameters calculated using the hydraulic samples with greater mean pore sizes, not in samples
conductivity model based on a lognormal pore size distribution, the
Kosugi model (Kosugi, 1999) with smaller pore sizes. The interpretation of the
actual effect of a on the hydraulic conductivity is
Mean Std. Dev. Max Min complicated by the fact that the term Sea does not
have an effect on the absolute magnitude of Ks, but
Mean pore 0.71 0.3 1.38 0.39
size (rm, mm)
only on the slope of the log(K ) –log(Se) plot. The
Mean capillary 23.5 6.71 38.2 10.9 figure indicates that the magnitude of a, representing
pressure (hm, mm) the slope of the log(K ) –log(Se) plot is governed by
Sigma (s ) 0.84 0.13 1.16 0.66 the mean pore size, with larger pore mean pore sizes
Alpha (a ) 20.06 0.65 1.28 21.69
Beta (b ) 0.84 1.22 2.78 21.07
having steeper slopes. As long as a is positive, this
means that the tortuosity decreases with increasing
R.W. Vervoort, S.R. Cattle / Journal of Hydrology 272 (2003) 36–49 45

Table 5 between the two parameters causes the b parameter to


Correlations between parameters in the Kosugi model (Kosugi, correct for positive slopes due to a (Table 5). In
1999) and measures of soil structural form, indicating the highest
practice, it means that a smaller a value (positive or
correlation between the tortuosity parameters and pore related
structural parameters such as MPIL and PSL negative) corresponds to a flat slope (small decrease in
tortuosity with increasing supply potential), which
Alpha (a ) Beta (b ) Mu (m ) Sigma (s ) can only happen if the width of the pore size
distribution (s ) is small. Although the relationship
Alpha (a ) 1.00 – – –
is not strong, there is a general positive relationship
Beta (b ) 20.64 1.00 – –
Mu (m) 0.63 20.49 1.00 – between s and a, indicating that smaller values of a
Sigma (s ) 0.50 20.59 0.84 1.00 were somewhat related to narrower pore size
Porosity 0.69 20.21 0.68 0.38 distributions (Table 5). The relationship between
SA 0.47 0.02 0.44 0.05 MPIL and a indicates that samples with a greater
MPIL 0.77 20.54 0.78 0.67
mean pore size tend to have greater values for a and,
MSIL 20.31 20.17 20.25 0.20
PSL 0.73 20.47 0.70 0.59 as such, a sharper increase in tortuosity with
SSL 20.32 20.16 20.29 0.12 decreasing pore size. Considering the fact that the
PG 0.71 20.40 0.80 0.59 conductivity is related to the fourth power of the pore
SG 20.03 0.42 20.15 20.50 radius, a greater mean pore size will tend to have a
greater Ks. The relationship developed here indicates
pore size (considering Eqs. (4) and (2)), and samples that this sample will also have a steeper hydraulic
with larger pores would have shorter path lengths. In conductivity curve (greater a ), and a wider pore size
other words, the average path length, or the distribution. This can be confirmed by the existence of
connectivity increases, as the sample desaturates a reasonable multi-linear relationship between Ks,
(Vogel, 1997). At MPIL values of around 1 mm, a MPIL and s (log(Ks) ¼ log(MPIL) þ s, r 2 ¼ 0:52).
becomes negative. Negative a values are related to So, even though a does not have a direct effect on the
positive slopes on the log(K )– log(Se) plot, which magnitude of Ks, it still seems that greater a values
would indicate higher conductivities at lower Se will correspond to greater values of Ks. Apart
values. This is of course not possible and the interaction from four samples, which were clearly outside

Fig. 3. The relationship between the mean pore intercept length (MPIL) and the a parameter from Kosugi (1999) model indicating an decrease
in tortuosity with larger MPIL.
46 R.W. Vervoort, S.R. Cattle / Journal of Hydrology 272 (2003) 36–49

Fig. 4. Relationship between a and log10 ðKs Þ indicating that larger a values correspond to larger values of Ks.

the relationship (open circles, these points were most of the samples there is a relationship between
incidentally all subsoil samples), this relationship pore related structural form measures and b (with
seems to exist in this dataset (Fig. 4). MPIL having the best relationship). The general
Little or no correlation existed directly between b relationship (Fig. 5) indicates an increasing tortuosity,
and the measures of soil structural form (Table 5). or decreasing connectivity (greater b values), with
However, inspection of the results indicates that for decreasing pore size. There are however three samples

Fig. 5. Relationship between mean pore intercept length (MPIL) and the b parameter in Kosugi (1999) model, indicating decreasing tortuosity
with increasing mean pore intercept length.
R.W. Vervoort, S.R. Cattle / Journal of Hydrology 272 (2003) 36–49 47

lying outside the general relationship. These three but there was no relationship between the earlier
samples all had negative b values around 2 1, calculated a and b parameters and this version of
compared to mainly positive or slightly negative tortuosity (averaged over all supply potentials). There
values in the other samples. There was only one other was also no correlation between this version of
sample, which had a negative b value close to 2 1. tortuosity and any of the soil structural form
Negative b values are possible if the pore size (r ) is parameters. Part of this could be due to the fact that
greater than unity (see Eq. (5)). Two of the samples this version of tortuosity includes the effect of the
with negative b values had indeed a mean pore size magnitude of Ks. However, no combination of Ks, a
(represented by MPIL) greater than unity. The two and b indicated any relationship with the tortuosity
other samples had lower mean pore sizes. The based on the conductivity ratio.
curvature of the log(K )–log(Se) plot for the samples The calculation of the tortuosity of a sample is
with negative b values is convex, in contrast to the apparently more complex than just the lack of fit from
mainly concave shape of all the other samples. In a capillary model. On the other hand, the parameters a
terms of tortuosity, a negative b value indicates an and b were both clearly correlated to pore related
increasing tortuosity with increasing pore size structural form, although none of the relationships
(Eq. (5)), similar to the negative a results. Positive was particularly strong. In contrast, the direct
b values are once again related to more logical relationship between soil structural form and Ks was
tortuosity values (increasing with decreasing pore much stronger. Part of this is caused by the fact that a
size). The negative b values were always related to and b are related to the relative hydraulic conductivity
positive a values (Table 5). and not to the magnitude of the saturated hydraulic
The structural form parameter PG, even though it conductivity. The few multi-linear relationships also
should represent the connectivity of the sample, never point at the complexity of the actual relationship. The
had a real strong correlation with either a or b. The analysis is further complicated by the theoretical
correlation between a and PG in Table 5 is dominated limitation that a should not be negative, while in
by a few samples with high PG values. The practice many of the curve-fitting results are negative.
relationship, however, seems to indicate that samples Part of this can be related to the correlation between
with more independent loops have an increasing the two parameters, which causes the slope of
tortuosity with increasing pore size (steeper slopes of the log(K )–log(Se) to be in the correct direction.
the log(K )–log(Se) plot, a is positive). SG was not However, the actual relationships of a and b and the
directly correlated to either of the parameters, but its mean pore size (MPIL) did indicate the expected
effect might be more complex. Using a multi-linear relationship with more porous samples with larger
model, the combination of porosity and SG has a pores having smaller tortuosities and thus higher
major effect on a (a ¼ log(porosity) þ log(SG), connectivities. But, considering the lack of really
r 2 ¼ 0:63), with a decreasing with increasing SG. clear relationships, the results presented here could
This indicates that for the more connected solid also indicate that:
phases, the difference in tortuosity between the largest
and smallest pores decreases (flatter slopes of the 1. the parameters in the hydraulic conductivity model
log(K )– log(Se) plot). have little physical significance (e.g. Vogel, 2000);
The ratio of the hydraulic conductivity based on 2. that the model is wrong;
straight capillary tubes to the observed hydraulic 3. that we are incapable of measuring the structural
conductivity (Jury et al., 1991) should also represent a features influencing the hydraulic conductivity; or
measure of tortuosity. This allows calculation of 4. that the parameters are a complex convolution of
tortuosities at each supply potential. Although there different effects.
was a wide variation in tortuosities (ranging from
1025 to 2 £ 1023), there was no significant variation Current thinking concentrates on the first issue,
in tortuosity with supply potential. This would with the parameters being treated as fitting variables
indicate similar path lengths at each supply potential. (the empirical approach) for the ‘soil structural
A few samples had higher than average tortuosities, effect’. This, however, limits our ability to predict
48 R.W. Vervoort, S.R. Cattle / Journal of Hydrology 272 (2003) 36–49

hydraulic conductivity across scales, or draw means the relationship between the genus numbers
conclusions about soil structure from the fitting and a and b is possibly complex.
results and it reduces the search for parameters to
predict hydraulic conductivities to random curve
fitting. Further complexity is introduced by the fact Acknowledgements
that the parameters in the model are not totally
independent. If the parameters have a clear physical We thank the Australian Cotton Cooperative
meaning, this would allow the derivation of logical Research Centre for funding this work.
estimates. The actual hydraulic conductivity model
could be wrong, and the recent efforts by Tuller et al.
(1999) and continued efforts in percolation modelling References
(e.g. Peat et al., 2000) might deliver new avenues in
this direction. This paper and the efforts by Perret et al. Anderson, A.L., McBratney, A.B., FitzPatrick, E.A., 1996. Soil
mass, surface and spectral fractal dimensions estimated from
(1999), Moreau et al. (1999) and Vogel (1997) try to
thin sections photographs. Soil Sci. Soc. Am. J. 60, 962 –969.
address the third issue, but much work still remains to Ankeny, M.D., Ahmed, M., Kaspar, T.C., Horton, R., 1991. Simple
be done. The fourth issue is limited by our technical field method for determining unsaturated hydraulic conduc-
abilities on this third issue and can also be an outcome tivity. Soil Sci. Soc. Am. J. 55, 467 –470.
of applying the new hydraulic conductivity models. Bear, J., 1972. Dynamics of Fluids in Porous Media, Elsevier, New
York.
However, without meaningful and quantitative
Berkowitz, B., Balberg, I., 1993. Applications of percolation theory
measures of soil structural form, we cannot decide in hydrology. Water Resour. Res. 29, 775– 794.
whether tortuosity and connectivity parameters in any Berryman, J.G., Blair, S.C., 1987. Kozeny– Carman relations and
model have a physical meaning, or whether other image processing methods for estimating Darcy’s constant.
effects than soil structure need to be taken into J. Appl. Phys. 57, 2374–2384.
Beven, K.J., Germann, P., 1982. Macropores and water flow in soils.
account. It is therefore important that continued effort Water Resour. Res. 18, 1311–1325.
is being put in methods to rapidly assess soil structural Broadbent, S.R., Hammersley, J.M., 1957. Percolation processes,
parameters, preferably in three dimensions. crystals and mazes. Proc. Cambridge Phil. Soc. 53, 629–641.
Cattle, S.R., Farrell, R.A., McBratney, A.B., Moran, C.J., Roesner,
E.A., Koppi, A.J., 2001. q Solicon—PC Version 2.1. The
University of Sydney and Cotton Research and Development
4. Conclusions Corporation.
Chen, M.H., Yan, P.-F., 1988. A fast algorithm to calculate the
Euler number for binary images. Pattern Recogn. Lett. 8,
Saturated hydraulic conductivity was well corre- 295 –297.
lated to several measures of structural form, in Childs, E.C., Collis-George, N., 1950. The permeability of porous
particular porosity and PG. The general model for materials. Proc. R. Soc. Lond. Ser. A 201, 392– 405.
hydraulic conductivity based on the lognormal pore Crawford, J.W., 1994. The relationship between structure and the
hydraulic conductivity of soil. Eur. J. Soil Sci. 45, 493 –502.
size distribution (Kosugi, 1999) was able to accurately
Fatt, I., 1956. The network model of porous media. 1. Capillary
predict the measured hydraulic conductivities, with pressure characteristics. Trans. Am. Inst. Min. Metall. Pet.
the range of a and b parameters indicating the range Engng 207, 144–181.
in connectivity and tortuosity between the different Fatt, I., Dijkstra, H., 1951. Relative permeability studies. Trans.
samples. The parameters a and b were also related to Am. Inst. Min. Metall. Pet. Engng 192, 249 –255.
Gardner, W.R., 1958. Some steady state solutions of the unsaturated
measures of soil structural form. The strongest moisture flow equation with applications to evaporation from a
relationship was between the MPIL and a and b. water table. Soil Sci. 85, 28–232.
This relationship indicated that greater MPIL values Giminez, D., Allmaras, R.R., Hugins, D.R., Nater, E.A., 1997.
were related to smaller b and greater a values. The Prediction of the saturated hydraulic conductivity-porosity
relationships confirmed the theoretical expectations. dependence using fractals. Soil Sci. Soc. Am. J. 61, 1285–1292.
Golden, J.M., 1980. Percolation theory and models of unsaturated
Pore and SG numbers, although measures of horizon- porous media. Water Resour. Res. 16, 201–209.
tal connectivity of the pore and solid phase, only had Isbell, R.F., 1996. The Australian Soil Classification, CSIRO
limited effect on the tortuosity parameters. This Publishing, Collingwood, Vic.
R.W. Vervoort, S.R. Cattle / Journal of Hydrology 272 (2003) 36–49 49

Jury, W.A., Gardner, W.R., Gardner, W.H., 1991. Soil Physics, Peat, D.M.W., Matthews, G.P., Worsfold, P.J., Jarvis, S.C., 2000.
Fifth ed, Wiley, New York. Simulation of water retention and hydraulic conductivity in soil
Kosugi, K., 1999. General model for unsaturated hydraulic using a three-dimensional network. Eur. J. Soil Sci. 51, 65–79.
conductivity for soils with lognormal pore-size distribution. Perret, J., Prasher, S.O., Kantzas, A., Langford, C., 1999. Three-
Soil Sci. Soc. Am. J. 63, 270–277. dimensional quantification of macropore networks in undis-
Kutilek, M., Nielsen, D.R., 1994. Soil Hydrology, Catena Verlag, turbed soil cores. Soil Sci. Soc. Am. J. 63, 1530–1543.
Cremlingen Destedt, 370p. Serra, J., 1982. Image Analysis and Mathematical Morphology,
Marshall, T.J., 1958. A relation between permeability and size Academic Press, London.
distribution of pores. J. Soil Sci. 9, 1–8. Soil Survey Staff, 1996. Keys to Soil Taxonomy, Seventh ed,
McBratney, A.B., Moran, C.J., 1990. A rapid method for Natural resource conservation service of the USDA, Washing-
analysis of soil macropore structure. II. Stereological model, ton, DC.
statistical analysis and interpretation. Soil Sci. Soc. Am. J. Tuller, M., Or, D., Dudley, L.M., 1999. Adsorption and capillary
54, 509 –515. condensation in porous media: liquid retention and interfacial
McBratney, A.B., Moran, C.J., Stewart, J.B., Cattle, S.R., Koppi, configurations in angular pores. Water Resour. Res. 35,
A.J., 1992. Modifications to a method of rapid assessment of 1949–1964.
soil macropore structure by image analysis. Geoderma 53, Vogel, H.J., 1997. Morphological determination of pore connec-
255–274. tivity as a function of pore size using serial section. Eur. J. Soil
Millington, R.J., Quirk, J.P., 1960. Permeability of porous solids. Sci. 48, 365–377.
Trans. Faraday Soc. 57, 1200–1207. Vogel, H.J., 2000. A numerical experiment on pore size, pore
Minasny, B., McBratney, A.B., 2000. Hydraulic conductivity connectivity, water retention, permeability and solute transport
pedotransfer functions for Australian soil. Aust. J. Soil Res. using network models. Eur. J. Soil Sci. 51, 99–105.
38, 905 –926. Vogel, H.J., Roth, K., 1998. A new approach for determining
Moreau, E., Velde, B., Terribile, F., 1999. Comparison of 2D and effective soil hydraulic functions. Eur. J. Soil Sci. 49, 547–556.
3D images of fractures in a Vertisol. Geoderma 92, 55 –72. Watson, K.W., Luxmoore, R.J., 1986. Estimating macroporosity in
Mualem, Y., 1976. A new model for predicting the hydraulic a forest watershed by use of a tension infiltrometer. Soil Sci.
conductivity of unsaturated porous media. Water Resour. Res. Soc. Am. J. 50, 578–582.
12, 2187–2193. Yanuka, M., Dullien, F.A.L., Elrick, D.E., 1986. Percolation
Mualem, Y., Dagan, G., 1978. Hydraulic conductivity of soils: processes and porous media. I. Geometrical and topological
unified approach to the statistical models. Soil Sci. Am. J. 42, model of porous media using a three-dimensional joint pore size
392–395. distribution. J. Colloid Interf. Sci. 112, 24 –41.

You might also like