Prediction of the penetrometer resistance of soils

with models with few parameters

W.R. Whalley
a,

, J. To
b
, B.D. Kay
b
, A.P. Whitmore
a
a
Rothamsted Research, West Common, Harpenden, AL5 2JQ, UK
b
Department of Land Resource Science, University of Guelph, Guelph, ON, Canada N1G 2W1
Received 14 March 2006; received in revised form 7 August 2006; accepted 29 August 2006
Available online 23 October 2006
Abstract
The objective of this paper is to determine to what extent pedotransfer functions, PTFs, can be developed that have few coefficients and which
are insensitive to soil type. The use of non-linear PTFs to predict penetrometer resistance of soils from their water status (matric potential, and
degree of saturation, S) and bulk density, , appears to require that some other soil property, such as sand content, is known. The use of a
logarithmic transformation on the dependent variable, Q and the independent variables, either or S has two effects. Firstly, it linearizes the data
and secondly it removes the increasing trend in the residuals of Q. A pedotransfer function derived from fitting log
10
Q to log
10
S and had
3 parameters that were insensitive to soil type. However, to predict Q on its natural scale, back-transformed values require correction for bias.
There is evidence that S is a better descriptor of soil water status than alone with respect to predicting penetrometer resistance. We show
that the use of S is preferable for both statistical and physically based reasons. However, we also show that matric potential can work well when
using PTFs to predict the strength of soil in the field given the variability in field measurements. We demonstrate how a PTF can be used to predict
values of the strength of field soil measured independently.
2006 Elsevier B.V. All rights reserved.
Keywords: Penetrometer resistance; Matric potential; Effective stress; Bulk density; Pedotransfer functions
1. Introduction
1.1. Penetrometer resistance and root growth pressure
The early use of penetrometers was frequently associated with
the problem of trafficability, but recently penetrometer resistance
has become an important measurement for assessing whether
roots can penetrate soil (Bengough and Mullins, 1990; Whalley
et al., 2005a; To and Kay, 2005). It is now understood that the
resistance of a soil to a penetrometer can be a factor of up to 3
times greater than the force which a root needs to exert to
penetrate the same soil (Bengough and Mullins, 1990). The
greater resistance of soil to a penetrometer in comparison with a
root can be explained largely by less frictional resistance between
roots and soil in comparison with metal and soil (Bengough and
McKenzie, 1997). Although lubricated penetrometers have been
shown to have lower penetrometer resistance (Tollner and Verma,
1987), a more practicable solution to take account of the soil to
metal friction when measuring penetrometer resistance is to rotate
the penetrometer (Bengough et al., 1997). Typically the fixed
penetrometer has a resistance to penetration two to three times
higher than that of a rotating penetrometer (Whalley et al., 2005a).
This is close to the ratio between the resistance to a fixed pene-
trometer and the pressure a root exerts to penetrate the same soil
(Bengough and Mullins, 1991). Analysis of the relationship bet-
ween penetrometer resistance and root growth pressure has in-
creased confidence in the use of penetrometers to assess the
rooting environment in soil. It is now generally accepted that a
fixed penetrometer resistance in excess of 2.5 MPa will seriously
restrict root elongation (e.g. Groenevelt et al., 2001).
1.2. A theoretical account of penetrometer resistance
To our knowledge the explanation of Farrell and Greacen
(1966) remains the clearest account of a theory describing the
relationship between soil properties and penetrometer resistance.
The different zones of failure that Farrell and Greacen proposed
Geoderma 137 (2007) 370377
www.elsevier.com/locate/geoderma

Corresponding author.
E-mail address: richard.whalley@bbsrc.ac.uk (W.R. Whalley).
0016-7061/$ - see front matter 2006 Elsevier B.V. All rights reserved.
doi:10.1016/j.geoderma.2006.08.029
around a penetrometer are shown in Fig. 1A. In the first of these
zones (i) soil is compressed to its minimum voids ratio (or
maximum bulk density), in the second zone (ii) the soil fails
plastically and in the outer zone (iii) the soil behaves as an elastic
material. Farrell and Greacen (1966) obtained an expression for
R in Fig. 1Aby equating the volume of voids before and after the
passage of the penetrometer. By solving the equations nume-
rically, they obtained the radial pressure needed to expand a
cavity in soil to accommodate the penetrometer. Although the
model of Farrell and Greacen (1966) was effective at describing
the penetrometer resistance (Fig. 1B), it is not useful in a practical
sense because it requires data fromthe compression characteristic
(Gregory et al., 2006) and the Mohr failure envelope which can
only be found from comprehensive testing of the soil.
1.3. Experimental observations of how soil properties affect
penetrometer resistance
Many accounts of the early use of penetrometers were
concerned with inferring soil properties from penetrometer
measurements of soil strength (e.g. Mulqueen et al., 1977) and
to some extent this interest continues today (e.g. Hernanz et al.,
2000). This work is instructive because it provides a detailed
account of the relationship between soil properties such as
shear, compressive and tensile strength and penetrometer
resistance. Of particular value is an analysis of how these soil
factors, which together determine penetrometer resistance, vary
with soil type, water content and bulk density. Mulqueen et al.
(1977) suggested that the mode of soil failure due to a pene-
trometer is a function of soil water content. At high water
contents, they suggested that the soil failed plastically and the
penetrometer resistance was insensitive to bulk density. At
intermediate water contents, soil failed by a combination of
compression and shear, and at low water contents, failure was
related to the internal friction of soil and was sensitive to bulk
density. They concluded that because of this complexity, it was
unlikely that penetrometer measurements could be used to infer
specific soil strength parameters. However, this does not affect
the main current use of penetrometer data to estimate the ease
Fig. 1. A. Soil failure zones ahead of a penetrometer being pushed into soil
reproduced from Farrell and Greacen (1966). B. A comparison of modelled and
measured penetrometer pressure. The modelled penetrometer pressure was that
obtained by Farrell and Greacen (1966) with a mechanistic model of a soil
failure ahead of a penetrometer.
Table 1
The organic carbon content (OC), bulk density (),. and the sand, silt and clay
contents of the undisturbed soil cores (see also To and Kay, 2005)
Site
#
Distribution Clay
b0.002 mm %
Silt 0.05
0.002 mm %
Sand 2.0
0.05 mm %
OC
%
D
b
Mg
m
3
1 Avg. 11.9 51.2 37.0 1.95 1.34
Min. 5.8 45.7 28.6 0.99 1.05
Max. 22.6 57.7 45.2 3.53 1.74
2 Avg. 13.3 50.3 36.4 2.00 1.37
Min. 8.3 41.9 25.9 0.99 1.05
Max. 22.6 58.7 46.4 3.42 1.74
3 Avg. 9.3 27.9 62.8 1.96 1.46
Min. 0.1 7.2 30.8 0.20 0.89
Max. 26.1 43.7 92.8 5.88 1.77
4 Avg. 44.9 44.6 10.5 2.40 1.37
Min. 34.6 40.3 2.6 1.27 1.06
Max. 53.0 48.0 25.1 3.35 1.68
5 Avg. 25.6 33.8 40.6 2.04 1.39
Min. 4.7 12.8 7.2 0.75 1.11
Max. 47.8 45.0 81.4 3.57 1.79
6 Avg. 36.4 43.5 20.1 2.17 1.41
Min. 24.4 20.4 10.8 0.87 1.07
Max. 47.9 55.1 53.9 3.64 1.74
7 Avg. 33.0 41.8 25.2 2.47 1.34
Min. 11.8 27.0 5.8 0.78 1.03
Max. 52.7 53.8 61.2 3.74 1.81
8 Avg. 17.1 54.5 28.3 1.87 1.52
Min. 7.1 39.4 7.1 0.39 1.22
Max. 47.1 64.5 53.5 3.03 2.07
9 Avg. 21.2 47.7 31.2 1.73 1.53
Min. 14.0 33.0 6.6 0.45 1.36
Max. 60.3 59.7 42.8 2.46 1.64
10 Avg. 16.9 55.3 27.8 1.80 1.48
Min. 9.6 37.9 19.2 0.43 1.28
Max. 27.5 68.1 46.5 2.85 1.75
11 Avg. 33.5 52.0 14.5 1.66 1.52
Min. 15.1 45.2 3.2 0.99 1.18
Max. 49.1 65.4 31.9 3.27 1.78
12 Avg. 17.8 32.2 50.1 1.75 1.55
Min. 5.9 11.7 28.4 0.83 1.35
Max. 32.4 43.5 80.1 2.93 1.80
371 W.R. Whalley et al. / Geoderma 137 (2007) 370377
with which roots can penetrate soil (see Section 1.1). The work
of Mulqueen et al. (1977) and others is useful because it can
help us to develop simple relationships between easily mea-
sured, logged or predicted soil properties (e.g. water content and
matric potential) and penetrometer resistance, and hence the
resistance to root penetration, which is more difficult to measure
and cannot be logged.
1.4. Predicting penetrometer resistance with pedotransfer
functions, PTFs
A key property of a useful PTF is that it must be applicable to
a wide range of soils and it must predict properties of soils that
were not included in the development data set. Recently To and
Kay (2005) described how a PTF to predict penetrometer
resistance, Q, could be derived from soil properties that change
with time (e.g. matric potential, water content and bulk density)
as well as those soil properties that are relatively constant with
time such as texture and organic matter content. The PTF that
fitted their data the best is written as
Q aw
b
cw 1
where is the matric potential and a, b, and c were functions of
texture, organic matter content and bulk density. For relatively
wet soils with low density
log
10
Q alog
10
r b 2
where (for low density soils) effective stress, , can be
approximated by
r Sw 3
in which S is the degree of saturation obtained by dividing soil
water content, , with the water content of the saturated soil (see
Whalley et al., 2005a). In Eq. (2), a and b were not sensitive to
soil type provided the bulk density of the soil was low.
The purpose of this paper is to use published data from
laboratory and field experiments to explore the extent to which
PTFs for soil penetrometer resistance can be developed that
have the minimum number of parameters and that do not
depend on soil type. We also explore the success of different
curve fitting approaches at predicting penetrometer resistance
from knowledge of soil water status and soil density.
2. Materials and methods
2.1. The laboratory data
To and Kay (2005) have collected a large set of data for
undisturbed soils and it is used here for curve fitting and model
testing. Briefly, the particle size classes, organic carbon content
and density of these soils are given by To and Kay (2005), but
reproduced here for ease of reference (Table 1). These soils
were collected in cores at depths of 5, 15 and 25 cm from 5 sites
in western Ontario in 1997 and 5 and 20 cm from seven sites in
1998. To and Kay (2005) report that all the sites were variable in
landscape and that they were planted with corn according to
tillage practices practised on the farms. Some of the sites were in
no-till for various lengths of time and others were in con-
ventional tillage consisting of autumn ploughing followed by
spring cultivation to make a seedbed.
Fig. 2. The penetrometer resistance of three soils (triangles: Parafield loam;
squares: Urrbrae loam and circles: Coleraine clay) packed to bulk densities from
1.0 to 1.7 cm
3
plotted against degree of saturation (A), matric potential (B) and
effective stress (C). These different measures of soil water status explained 0,
54% (P=0.004) and 70% (Pb0.001) of the variance in penetrometer resistance
respectively. This is a reanalysis of the data of Farrell and Greacen (1966) in
Whalley et al. (2005b).
372 W.R. Whalley et al. / Geoderma 137 (2007) 370377
A cone penetrometer with a 30 cone angle and a basal
diameter of 4 mm was used to make penetrometer resistance
measurements. The rate of penetration was 2 mm/min. The
penetrometer resistance of the undisturbed cores was measured
following equilibration at matric potentials of 1, 3, 6, 10,
33, 100 and 1500 kPa. In total, penetration resistance was
determined on 717 cores. After the penetrometer resistance of
each core was measured and the soil water content was
determined. Bulk density, texture and organic carbon contents
are shown in Table 1.
We chose to use the data from undisturbed soils to avoid
artefacts that can result when cores are packed to a high density
and to ensure as far as is possible that the combinations of
density, texture and organic carbon contents reflect those that
are likely to occur in a managed agricultural system.
We also used penetrometer resistance data from undisturbed
cores at Roseworthy College, South Australia, reported by
Groenevelt et al. (2001). This was a non-swelling soil and the
resistance to penetration was measured with a 2 mm diameter
cone with a penetration rate of 2 mm min
1
on soil equilibrated
at 1, 3, 5, 10, 33, 55, 100, 500 and 1500 kPa.
Similar data presented by Farrell and Greacen (1966) are also
used in this paper. They measured the penetrometer resistance of
three soils repacked to bulk densities between 1.0 and 1.7 g cm
3
which were equilibrated at matric potentials of 30 at 70 kPa.
2.2. Data from field experiments
Whalley et al. (2006a) measured the penetrometer resistance
of loamy sand during the spring and early summer of 2004 in a
field experiment to investigate the effects of soil drying on soil
strength. Winter wheat was grown and the soil was either well-
watered to keep it at a matric potential of 5 kPa at a depth of
20 cm, prevented from drying to less than 80 kPa, or it was not
irrigated. The matric potentials were monitored at a depth of
20 cm with a combination of water-filled tensiometers and
porous matrix sensors (Whalley et al., in press). The soil water
content was also monitored at a depth of 20 cm with a dielectric
soil moisture meter (Whalley et al., 2004).
3. Results and discussion
3.1. Appropriate measures of soil water status to predict
penetrometer resistance
Estimates of soil water status and soil density are both
needed to predict Q in soil. Soil water status can be defined
by , S, , or the product S. Effective stress can be used to
predict the tensile strength of soils with relationships that can
be applied successfully to a wide range of soils provided S is
greater than 0.5 (Mullins, 2000). The deformation of soil by a
penetrometer is more complex than simple tensile failure, but
nevertheless, Whalley et al. (2005a) have shown that effective
stress can be used in a single expression (see Eq. (2)) to
provide a common prediction of the penetrometer resistance
of soils with a low bulk density. The most striking benefit of
the use of effective stress is to be seen from re-examination of
the data of Farrell and Greacen (1966), reproduced in Fig. 2.
The use of effective stress gave a better prediction of
penetrometer resistance than matric potential, but the degree of
saturation was not correlated with penetrometer resistance. For
the soils studied by Farrell and Greacen (1966) the use of
effective stress gave a good prediction of penetrometer
resistance over a range of soil bulk densities, but this is an
exception. The original use of effective stress, , for soil in
equilibrium with atmospheric pressure is described by Bishop's
equation (Bishop and Blight, 1963)
r Pvw 4
where P is the total stress on a soil at a matric potential, and
at a degree of saturation S. Here is a factor that takes into
account the number of water-filled menisci at . For tensile
failure when SN0.5, can be approximated by S (Mullins,
2000; Whalley et al., 2005a). In very loose soils, strength is
determined almost entirely by soil water status; P is small so
the strength of the soil is determined by the product S. Under
these conditions effective stress, S. This explains why
the penetrometer pressure of low density soil can be predicted
from S by a single equation (e.g. Eq. (2)) with the same
fitted coefficients for a wide range of soils. In dense soils,
when total stress, P, is high the effective stress, , which
characterizes the intrinsic strength of soil, depends on both S
which is the component of strength due to soil water status as
Fig. 3. A plot of organic carbon against the clay content of the soils presented in
undisturbed soils listed in Table 1. The boundaries indicate a section of data for
soils with a range of organic carbon contents up to 5% which are not correlated
with bulk density.
Table 2a
The linear and non-linear models fitted to penetrometer resistance, Q (kPa),
where
w
(kPa) is effective stress, % sand is the percentage of sand in the
samples, (g cm
3
) is the bulk density, (kPa) is the moisture potential and a,
b, c and d are constants
log
10
Q=alog
10

w
+b+c (8)
log
10
Q=alog
10
+b+c(OC) +d (9)
log
10
Q=alog
10

w
+b+c

%sand+d (10)
Q=
w
b+c
(11)
Q=
w
b+c%sand+d
(12)
Q=
b+c
(13)
Q=
b+c%sand+d
(14)
373 W.R. Whalley et al. / Geoderma 137 (2007) 370377
well as P. For clarity we will define
w
=S to identify it as
the component of soil strength (or total stress) that is due to
the soil water status.
The use of soil water content to predict penetrometer
resistance (e.g. Grumwald et al., 2001) is likely to result in
functions that have a greater dependence on soil type than
would be the case if either or
w
were to be used. However, in
many cases soil water content may be the only available
measure of soil water status.
3.2. Effect of density and organic carbon content on the
penetrometer resistance of low density soils
The data set of To and Kay (2005) for undisturbed soils
(Fig. 3) show the widely reported decrease in bulk density with
increasing organic carbon content. In a subset of these data,
when bulk density, , is less than 1.2 g/cm
3
and organic carbon
content, OC, less than 5% these two soil properties are not
correlated significantly ( p=0.62). To predict the penetrometer
resistance of this subset of the data from soil moisture status the
following regression equations were obtained
log
10
Q 0:27log
10
w 2:38q 0:14OC 5
or
log
10
Q 0:32log
10
r
w
2:01q 6
which explained 63.2 and 68.5% of the variance in log
10
Q
respectively. In Eq. (6) based on
w
, the inclusion of organic
carbon content did not significantly improve the fit to the data
( p=0.112). The significance of Eq. (5) based on was greatly
reduced without the terms in OC and (Tables 2a and b). This
suggests that the effect of organic carbon on the penetrometer
resistance of these low density soils is related to its effects on the
water release characteristic, which is taken into account when
log
10
Q is expressed as a function of log
10

w
, but not when
expressed as a function of log
10
. We note that, for these low
density soils in both Eqs. (5) and (6), the constant term was not
significantly different from zero.
3.3. Penetrometer resistance as a function of soil water status
and soil density
For low density soils the function obtained by Whalley et al.
(2005a)
log
10
Q 0:76log
10
r log
10
62 7
gives good agreement with data from the To and Kay (2005)
data set for soils with a bulk density less than 1.2 g/cm
3
(Fig. 4)
and also those data reported by Groenevelt et al. (2001) and
Farrell and Greacen (1966). Here we note that
w
. However,
these bulk densities are unrealistically low for most situations.
With the data of To and Kay (2005), for undisturbed soil, the
simple model of Whalley et al. (2005a) has been extended to
include soil bulk density and the new model can be written as
log
10
Q 0:35F0:009log
10
r
w
0:93F0:0572q
1:2623F0:0832 8
This explains 74.8% of the variance in log
10
Q (Fig. 5) and
the standard errors of the fitted parameters are given in brackets.
Table 2b
To test the value of using extra parameters in the models in Table 2a, such as those for the sand content, the F statistic is calculated as follows: RSS/df/RMS of the
more complex model where RSS is the change in residual sum of squares df is the change in degrees of freedom and RMS is the residual mean square of the more
complex model
Model df RSS
a
RMS
a
Tested against F Result Mean square error
(8) 553 20.58 0.0372 (9) 8.71

(8) is better 0.193

(9) 552 20.91 0.0378 0.195
(10) 552 20.33 0.0368 (8) 6.79

(10) is better 0.192

(11) 553 354,700,000 641,500 (13) 1.002NS (11) not worse 800
(12) 552 273,800,000 496,100 (11) 163

(12) is better 704

(13) 553 355,600,000 643,100 801
(14) 552 317,300,000 574,900 (13) 66.56

(14) is better 758

NS not significant.
a
Four significant figures are presented.
pb0.05.
pb0.01.
pb0.001.
Fig. 4. Prediction of log
10
Q from effective stress using the relationship for low
density soils developed by Whalley et al. (2005a) (solid line). Data plotted are
those published by To and Kay (2005) for a bulk densities less than1.2 g cm
3
and matric potentials of 1.5 MPa or less (). The data published by Groenevelt
et al. (2001) () and Farrell and Greacen (1966) () are also plotted.
374 W.R. Whalley et al. / Geoderma 137 (2007) 370377
If log
10
is used to explain the variation in log
10
Q, both soil
density and organic carbon content (Eq. (9) in Table 2a) were
needed in the regression to explain 74.3% of the variance in
log
10
Q. These extra variables decrease the number of degrees
of freedom available to test the residual mean square and
decrease the significance of the relationship (Table 2b).
The relationships between Q and soil water status and
density are non-linear. However, they can be linearized by log
transformation (Eqs. (8)(10) in Table 2a). Log transformation
before fitting is statistically desirable because it eliminates the
increase in the variance of Q with increasing Q. However, the
prediction of log
10
Q rather than Q may be less appealing. In
Table 2a we have listed two linear models and their non-linear
counterparts to relate Q to
w
, , and in one case also the sand
content of the soil. A non-linear model to predict Q from matric
potential and soil bulk density is also listed. Including the sand
content of the samples gives a significant improvement in all
cases but the benefit is much greater in the two non-linear
models (12) and (14). There is a benefit in using
w
instead of,
since the smaller RMS with models (11) and (12) indicates
that it is better to use these than the corresponding models (13)
and (14). When limited data on soil properties is available, we
recommend that model (8) in Table 2a is used to predict log
10
Q
because it has parameters that are not sensitive to soil type (i.e. it
does not contain OC or sand content), and its inputs are either
easily measured with sensors (i.e. S and ) or they depend on
properties such as bulk density that change slowly in rigid soils
or that can be calculated from shrinkage in soils with a high clay
content. When more data on the soil is available such as texture
and organic carbon content then the functions such as model (10
or 12) or those described by To and Kay (2005) can be used to
estimate Q.
Transforming to logarithm enabled us to fit a model to data
where the variance in the independent variable increases with the
dependent variable(s), here stress
w
or and . The residuals
derived in this way have no trend whereas the residuals on the
natural scale increase with
w
or . Although the parameters in
Eq. (8) are therefore unbiased, the relationship is not helpful
where there is a need to predict Q rather than log
10
Q. In
transforming back, bias is introduced and in general the
predicted values will be too low. The simplest way to avoid
this is to multiply the back transformed value (i.e taking the anti-
logarithm log
10
Q predicted in Eq. (8)), by the mean of the anti-
logarithm of the residuals,
1
n
X
10
QQ

15
where Q

is the fitted value. Using the data of To and Kay (2005)

this factor has a value of 1.097.
Strictly, multiplication of Eq. (8) with Eq. (15) is valid only
where there is no trend in the residuals, which is not the case
here. A better correction can be applied by taking account of
how the residuals vary with
w
and but this may not work
well where there is little or no replication or if the relationship
is intended for extrapolation outside the range used to derive it.
Expression (15) is much more robust under these circum-
stances even though some bias will remain. Essentially Eq.
(15) is an empirically-derived correction factor that is widely
applicable to soils similar to the ones studied here. A PTF
based on its use will be improved by the collection of more
data as will Eq. (8). The combined use of Eqs. (8) and (15) to
predict Q provides a PTF with only 3 parameters all of which
are insensitive to soil type.
3.4. Predicting the strength of soil in the field
The discussion to this point has been concerned with the use
of PTFs to predict the strength of soils in cores. However, for
these to be useful they must also give a prediction of the
penetrometer resistance in the field. The field environment is
more complicated than that of soil cores because, even for a soil
profile with a uniform water content and density at all depths,
penetrometer resistance will tend to increase with depth because
of the difficulty in moving soil particles that support the weight
of the soil above (Fig. 6). However, the strength of the surface
Fig. 5. Using the complete data set of To and Kay (2005) log
10
Q is plotted
against log
10
and bulk density. The plane corresponds to Eq. (8) and the line
plotted is that for low density soils derived by Whalley et al. (2005a) (Eq. (2)).
Fig. 6. Penetrometer resistances of non-compacted plots in 2004 (Whalley et al.,
2006a). The increase in penetrometer resistance with time is attributed to soil
drying and the increase in penetrometer resistance with depth is related to both
soil drying and increasing surcharge (the increase in confining pressure with
depth due to the weight of soil).
375 W.R. Whalley et al. / Geoderma 137 (2007) 370377
layer soil is almost entirely due to its density and water status.
The penetrometer resistance of soil measured at a depth of
20 cm in a field experiment (see Whalley et al., 2006a) has been
predicted with Eq. (8) which was derived from the independent
data of To and Kay (2005) and also with their PTF (Eq. (1)).
The combined use of Eqs. (8) and (15) has the advantage that
prediction of Q can be made with a model that has only
3 parameters none of which is dependent on soil type. The
prediction of Q with this approach differs by 0.039 MPa on
average from the measured values with an SE of 0.039 and this
error is not significantly different from zero. To and Kay's
(2005) model differs by 0.22 MPa on average with an SE of
0.045 but this small difference does differ significantly from
zero (Fig. 7).
It should be noted that the measured penetrometer data
shown in Fig. 7 was obtained from soils with values of S greater
than 0.5 with the exception of two cases when S was close to
0.45. If S becomes very small it is likely that the To and Kay
(2005) PTF would give a better prediction of Q. However, these
very dry soils are likely to be too strong for any root elongation.
3.5. Soil-type dependence in PTFs
The PTF developed by To and Kay (2005) (see Eq. (1)) has
parameters that depend on soil texture, organic matter content
and bulk density. Predictions of Q can then be made for various
soils as a function of for different soils using the appropriate
set of parameters. In Eq. (8), Q is simply a function of and .
Although we have shown that this simpler PTF can work well
for an independent data set obtained with sandy soil (Fig. 7), the
extent to which it can account for soil-type dependence in the
relationship between Q and merits discussion. In Eq. (8) for a
given value of , there will be a spread in values of
w
because
S will be different for different soils according to the water
release characteristic. In Fig. 8, the fitted values of log
10
Q and
the back transformed estimates of Q are plotted against
representative measured data for clay and sandy soils. This
shows that the variation in Q with soil type can be accounted for
in PTFs that are based simply on the use of
w
and bulk density.
3.6. Uncertainty analysis
Eq. (8) is a relatively simple function, so that amplification of
input errors is unlikely in its use. Nevertheless we examined the
effects of combinations of likely values of the inputs (bulk
density, 1.21.65;
w
25200 kPa) on the predicted value of
log
10
Q using Eq. (8). These results are summarised in Table 3.
The resultant variance in log
10
Q is close to the input variance in
bulk density and less than the variance in
w
emphasising the
importance of precise estimates of the former. Eq. (8) does not
appear to amplify variation during the propagation of input error
into output. However, and
w
may be weakly correlated in
Fig. 7. Prediction of the penetrometer resistance at a depth of 20 cm in the field
during 2004 (see Whalley et al., 2006a). Predictions by the combined use of Eqs.
(8) and (15) () in this paper and by the pedotransfer function of To and Kay
(2005) () are plotted against measured data. It is important to note that both of
these pedotransfer functions were developed using the data of To and Kay
(2005) which is completely independent of the soil from which these
measurements were obtained (Whalley et al., in press).
Fig. 8. Fitted values of log
10
Q and Q obtained with Eqs. (8) and (15) plotted
against the measured data. Here only 2 subsets of the data used in the model
development are presented. These subsets represent sandy soils () (with a sand
content greater than 70%) and clay soils () (with clay content greater than
45%).
Table 3
The results of a Monte-Carlo randomisation of the inputs ( and
w
) to
investigate the propagation of errors in Eq. (8) (uncertainty analysis)
Inputs Output
log
10

w
log
10
Q
Mean 1.425 1.96 3.28
Median 1.425 2.05 3.28
Variance 0.0209 0.0941 0.0297
376 W.R. Whalley et al. / Geoderma 137 (2007) 370377
practice and if so this might alter the output variance slightly.
The data in Table 3 were obtained by full Monte-Carlo rando-
misation of the inputs.
4. Conclusions
If little is known about the soil, then log
10
Q can be predicted
with Eq. (8), provided both soil density and soil water status are
known. We have shown how to correct for bias in estimates of
Q obtained by back-transformation. However, an estimate of the
mean of the anti-logarithm of the residual error is needed,
although Eq. (15) is likely to be widely applicable. If soil texture
and organic matter content are known, non-linear functions can
be used to predict Q on its natural scale. Although we have
shown that
w
is preferable to the use of matric potential,
because it either describes more of the variance or requires the
use of fewer parameters, Q can be predicted from if a
reduction in precision is acceptable or if additional information
on the texture and soil organic carbon content is available.
Acknowledgements
Rothamsted Research is grant-aided by the Biotechnology
and Biological Sciences Research Council. We thank Dr. C.E.
Mullins for helpful comments on an earlier draft of this paper.
References
Bengough, A.G., McKenzie, B.M., 1997. Sloughing of root cap cells decreases
the frictional resistance to maize (Zea mays L.) root growth. J. Exp. Bot. 48,
885893.
Bengough, A.G., Mullins, C.E., 1990. Mechanical impedance to root growth: a
reviewof experimental techniques and root growth responses. J. Soil Sci. 41,
341358.
Bengough, A.G., Mullins, C.E., 1991. Penetrometer resistance, root penetration
resistance and root elongation rate in 2 sandy loam soils. Plant Soil 131,
5966.
Bengough, A.G., Mullins, C.E., Wilson, G., 1997. Estimating soil frictional
resistance to metal probes and its relevance to the penetration of soil by
roots. Eur. J. Soil Sci. 48, 603612.
Bishop, A.W., Blight, G.E., 1963. Some aspects of effectives stress in saturated
and partly saturated soils. Geothechnique 13, 177197.
Farrell, D.A., Greacen, E.L., 1966. Resistance to penetration of fine probes in
compressible soil. Aust. J. Soil Res. 4, 117.
Gregory, A.S., Whalley, W.R., Watts, C.W., Bird, N.R.A., Hallett, P.D.,
Whitmore, A.P., 2006. Calculation of the compression index and the
precompression stress from soil compression test data. Soil Tillage Res. 89,
4557.
Groenevelt, P.H., Grant, C.D., Semesta, S., 2001. A new procedure to determine
water availability. Aust. J. Soil Res. 39, 577598.
Grumwald, S., Rooney, D.J., McSweeny, K., Lowery, B., 2001. Development of
pedotransfer functions for a profile cone penetrometer. Geoderma 100, 2447.
Hernanz, J.L., Peixoto, H., Cerisola, C., Sanchez-Geron, V., 2000. An empirical
model to predict soil bulk density profiles in field conditions using penetration
resistance, moisture content and soil depth. J. Terramechs. 37, 167184.
Mullins, C.E., 2000. Hardsetting soils. In: Sumner, M. (Ed.), Handbook of Soil
Science. CRC Press, pp. G65G85.
Mulqueen, J., Stafford, J.V., Tanner, D.W., 1977. Evaluation of pentrometers for
measuring soil strength. J. Terramechs. 14, 137151.
To, J., Kay, B.D., 2005. Variation in penetrometer resistance with soil properties:
the contribution of effective stress and implications for pedotransfer functions.
Geoderma 126, 161276.
Tollner, E.W., Verma, B.V., 1987. Lubricated and non-lubricated cone penetrom-
eter performance comparison in six soils. Trans. ASAE 30, 16111618.
Whalley, W.R., Cope, R.E., Nicholl, C.J., Whitmore, A.P., 2004. In-field
calibration of a dielectric moisture meter designed for use in access tube. Soil
Use Manage. 20, 203206.
Whalley, W.R., Leeds-Harrison, P.B., Clark, L.J., Gowing, D.J.G., 2005a. The
use of effective stress to predict the penetrometer resistance of unsaturated
soils. Soil Tillage Res. 84, 1827.
Whalley, W.R., Clark, L.J., Gowing, D.J.G., Leeds-Harrison, P.B., 2005b. Soil
strength as a constraint to the growth of wheat. Asp. Appl. Biol 73, 1926.
Whalley, W.R., Clark, L.J., Gowing, D.J.G., Cope, R.E., Lodge, R.J., Leeds-
Harrison, P.B., 2006a. Does soil strength play a role in wheat yield losses
caused by soil drying? Plant Soil 280, 279290.
Whalley, W.R., Clark, L.J., Take, W.A, Bird, N.R.A., Leech, P.K., Cope, R.E.,
Watts C.W., in press. A porous-matrix sensor to measure the matric potential
of soil water in the field. European Journal of Soil Science. (Available online
doi:10.1111/j.1365-2389.2006.00790.x).
377 W.R. Whalley et al. / Geoderma 137 (2007) 370377