Professional Documents
Culture Documents
A R T I C L E I N F O
A B S T R A C T
Article history:
Received 20 May 2011
Received in revised form 19 March 2012
Accepted 14 May 2012
Soil compaction is a form of physical degradation that causes soil densication and distortion that can
lead to changes in three-dimensional soil structure variability. For this reason we proposed a method for
mapping compaction in three-dimensional space (3D), based on simple soil-specic depth functions.
This method, denoted the top down method, is based on a multi-step approach that starts with a
framework for selecting the best depth function, considers which shape best describes the soil-specic
variation by depth and ends with an interpolation of the coefcients of the function across the eld.
Subsequently, for each interpolated point this function is solved in order to estimate compaction in 3D
space. The top down method was evaluated using a cone resistance dataset, collected at the CULS (Czech
University of Life Sciences) farm in Lany. The accuracy of this method in predicting cone-index variation
in the 3D space was compared with the generally accepted 3D ordinary kriging interpolation, using
descriptive statistics of the predicted values and cross validation. We found that the top down method
better represented observed information and generally performs better than 3D kriging with a far
smaller sample size. The cross validation results suggest that ordinary kriging can predict cone-index
obtaining a value of root mean squared deviation (RMSD) of 0.54 at 10% soil proles excluded, and 0.57 at
50% exclusion; on the contrary the top down mapping method obtains an RMSD value of just 0.47, at 10%
exclusion, and 0.53, with 50% exclusion. We also found that the depth function method was more
representative of the observed variability in the predicted dataset, avoiding the smoothing ltering
effect typical of kriging interpolation. We then created a cone-index map of the CULS Lany eld with a
vertical resolution of 10 cm using the top down method. The map can be seen as a series of three videos
where it is possible to observe the cone-index pattern in three-dimensions.
2012 Elsevier B.V. All rights reserved.
Keywords:
Soil cone-index
Three dimensional digital soil mapping
Soil compaction
Kriging
1. Introduction
Soil compaction is a form of physical degradation that causes
soil densication and distortion decreasing soil porosity and
permeability. This obstructs the movement of air, water and
nutrients and restricts root penetration. In arable soils compaction
is often caused by repeated ploughing or heavy harvesting
machinery and it can occur at the surface or in subsurface
horizons. Ploughing can rectify topsoil compaction but once
subsoil compaction occurs it is extremely difcult and expensive to
alleviate. Compaction may also accumulate over time, creating
compacted soil layers that are almost impermeable and highly
resistant to root penetration (Jones and Montanarella, 2001), i.e. a
plough pan. This changes the three-dimensional soil structure
variability, and these modications can be high even in relatively
homogeneous soil (Carrara et al., 2007). Therefore we require a
rapid and accurate method for mapping soil compaction and the
changes in three-dimensional soil structure that it causes.
* Corresponding author.
E-mail address: f.veronesi@gmail.com (F. Veronesi).
0167-1987/$ see front matter 2012 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.still.2012.05.009
112
Table 1
Brief statistical description of the cone-index dataset used in the study.
Cone-index data (MPa)
Mean
Standard deviation
Median
Minimum value
Maximum value
Skewness
Kurtosis
Octile skew
Variance
1.95
0.73
1.88
0.82
4.2
0.49
0.65
0.55
0.54
Fig. 1. Location of the study area. The CULS Lany farm is in the Stredocesky Region, near the city of Ruda, about 40 km W of Prague.
113
Fig. 2. (a) Plot cone-index/depth for the entire Lany dataset. In this graph is possible to appreciate the general pattern of the dataset. (b) This pattern is not recognizable in
every soil prole but the Eq. (1) is perfectly able to describe both proles.
(1)
For every point of depth x(i,j), the value of the soil compaction,
y(i,j), can be calculated with a polynomial in which b1, b2 and b3 are
the coefcients of the equation, tted with a least squares
estimation.
In order to examine in detail the behaviour of the soil
property and the error associated with the regression,
we present a spatial study of the goodness of t measures
(e.g. RMSD) and of the residuals from the polynomial
regression.
2.3. Geostatistical interpolation
The top-down method relies on the tting of a polynomial
depth function to the cone-index data plus an interpolation of the
coefcients of the polynomial. We used an ordinary kriging
interpolation, with the variogram model tted using a residual
maximum likelihood (REML) estimator. We choose the REML
estimator because it was demonstrated by Kerry and Oliver
(2007) that prediction based on variogram models tted with the
REML estimator are generally more accurate and a sample size of
around 50 sites is adequate for kriging under these circumstances.
In order to demonstrate the accuracy of our method, we
compare it with the generally accepted 3D ordinary kriging, which
has already been used by Castrignano` et al. (2002) for mapping
cone-index data.
We refer to Deutsch and Journel (1998), Lark et al. (2006), Kerry
and Oliver (2007) and Webster and Oliver (2007) for a detailed
description of the two methods.
114
2.4. Validation
The validation process involved a random subsampling of a
percentage of soil proles (10, 20, 30, 40 and 50% of the population).
We validate the mapping method by re-predicting the cone-index
values of the subsampled population, and by comparing the
observed value with the predicted one using common statistical
indexes, e.g. R2 and RMSD, repeating the process 500 times.
2.5. Software
The geostatistical analyses were undertaken with the statistical
programming language R (R Development Core Team, 2009) and
the package gstat (Pebesma, 2004). The three dimensional
visualization of the soil map was created using Golden Software
Voxler (Golden Software, 2008).
3. Results and discussion
3.1. Polynomial depth function
We developed the topdown mapping for describing the
horizontal and vertical variation of cone resistance, based on 3D
mapping with a simple soil-specic depth function. This method is
based on the assumption that penetrometer data are characterized
by uncertainties at each measured point due in part to the use of a
manually inserted instrument that is unlikely to be inserted into
the soil at a constant speed each time. For this reason our
hypothesis is that a simple depth function can optimally describe
the general cone resistance prole without being constrained by
these point uncertainties that would cause errors with more
complex functions. The topdown mapping is based on a
framework developed for determining the best site specic depth
function, from which we selected the polynomial in Eq. (1) as best
predictor. This function can describe the average pattern (Fig. 2a)
very accurately, because its shape effectively reproduces a pattern
115
Fig. 4. (a) and (b) Interpolated maps, with a spline algorithm, of the two goodness of t indexes: R squared and RMSD. The two colour scales are created in a way that light grey
reects areas of good performances and darker colours indicate areas of worst performance. (c) Map of the residuals for the Eq. (1). The colour scale is centre based, therefore
also in this case a darker colour reect a worse performance.
Table 2
Results of the cross validation.
3d Ordinary kriging
Exclusion percentage
10
20
30
40
50
10
20
30
40
50
Mean
Median
Standard deviation
Minimum
Maximum
0.54
0.53
0.10
0.32
0.81
0.54
0.54
0.07
0.36
0.77
0.55
0.56
0.05
0.41
0.70
0.56
0.56
0.05
0.44
0.70
0.57
0.57
0.04
0.47
0.68
0.47
0.46
0.10
0.28
0.99
0.48
0.47
0.09
0.32
0.84
0.49
0.48
0.07
0.34
0.84
0.52
0.48
0.32
0.36
6.18
0.53
0.49
0.45
0.38
9.14
Fig. 5. Results of the independent validation. The accuracy of the proposed mapping
method is higher than 3D kriging even when the percentage of exclusion reaches
50%, meaning that it can theoretically be used for mapping with a fraction of the
original dataset.
116
Fig. 6. Results of the comparison. Three depth layers are presented: 0.1 m (surface), 0.3 m (approximate depth of the plough layer) and 0.5 m (maximum investigation
depth). On the left are depicted the results of the 3D kriging and on the right the results of the proposed mapping method.
Busscher et al., 1997; Hamza and Anderson, 2005). At the CULS Lany
farm, the variation in the cone-index values, especially in the topsoil,
is predominantly caused by land use. The eld is covered by arable
crops and is managed by conventional ploughing technology. The
predominance of the land use over other factors is illustrated by the
relative homogeneity of the topsoil. There is a transition zone, at
around 30 cm, between the managed topsoil and the natural subsoil,
and this is observed as an increase in the cone resistance values.
However, this increase of the soil compaction is not always related to
the presence of a plough pan, in certain areas the increase is
smoother. This pattern could suggest the presence of an external
phenomenon that has modied the original soil pattern at that
depth, possibly a paleobed which would be compatible with the
alluvial deposits below the study eld.
Below 35 cm the soil pattern becomes more complex with areas
of high cone-index data and areas with low cone-index values
distributed without any recognizable pattern. In general, directly
below the plough layer, there is a decreasing tendency in the coneindex values but there are areas in which high values of cone-index
are maintained throughout the survey depth. Although the
homogeneity of the topsoil is certainly due to the Lany land
management, the factors that may cause the observed patterns in
the subsurface to be less easily identiable and compaction may be
derived from several factors. There are not enough textural and
moisture data to fully clarify which one is the cause of the depth
distribution of the cone-index data below the plough layer.
117
Fig. 7. 3D maps of Lany eld. These maps are the results of the mapping procedure proposed in the text. They are created by mixing two kinds of 3D visualization, 2D slices and
contours. Each image is composed by ve slices on the XZ plane, spaced by 100 m, ve slices on the YZ plane, spaced by 100 m, and one contour slice, on the XY plane, at the
three depth layers indicated (0.07, 0.30 and 0.50 m).
4. Conclusions
We present a novel method, based on site specic depth
functions, to improve the accuracy of predicting cone resistance
variation by depth. With this technique we can decrease the
impact of the unknown uncertainties related to penetrometer
measurements. We also demonstrate with descriptive statistics
and cross validation that the proposed method is more accurate
than 3D kriging, and we suggest that its results are easier to
interpret.
This study presents an example of soil mapping with a limited
dataset and its successful application to 3D mapping. Although this
study was limited to a eld with particular soil properties, in
118
Busscher, W.J., Bauer, P.J., Camp, C.R., Sojka, R.E., 1997. Correction of cone index for
soil water content differences in a coastal plain soil. Soil & Tillage Research 43,
205217.
Carrara, M., Castrignano`, A., Comparetti, A., Febo, P., Orlando, S., 2007. Mapping of
penetrometer resistance in relation to tractor trafc using multivariate geostatistics. Geoderma 142, 294307.
Castrignano`, A., Maiorana, M., Fornaro, F., Lopez, N., 2002. 3D spatial variability of
soil strength and its change over time in a durum wheat eld in Southern Italy.
Soil & Tillage Research 65, 95108.
Czech Geological Survey, 2004. Interactive Geological Map. http://mapy.geology.cz/
website/mpr_an/vi.pl.
Deutsch, C.V., Journel, A.G., 1998. GSLIB: Geostatistical Software Library and Users
Guide. Oxford University Press.
Golden Software, 2008. Voxler Software for 3D Data Visualization. http://
www.goldensoftware.com/.
Hamza, M.A., Anderson, W.K., 2005. Soil compaction in cropping systems: a review
of the nature, causes and possible solutions. Soil & Tillage Research 82, 121145.
Hiederer, R., 2009. Distribution of Organic Carbon in Soil Prole Data. EUR 23980
EN. Ofce for Ofcial Publications of the European Communities, Luxembourg,
p. 126.
Jones, R.J.A., Montanarella, L., 2001. Subsoil compaction: a hidden form of soil
sealing in Europe. In: O-COPERNICUS, 3rd workshop, Busteni, Romania.
Kerry, R., Oliver, M.A., 2007. Comparing sampling needs for variograms of soil
properties computed by the method of moments and residual maximum
likelihood. Geoderma 140, 383396.
Lark, R.M., Cullis, B.R., Welham, S.J., 2006. On spatial prediction of soil properties in
the presence of a spatial trend: The empirical best linear unbiased predictor (EBLUP) with REML. European Journal of Soil Science 57, 787799.
Lark, R.M., Bishop, T.F.A., 2007. Cokriging particle size fractions of the soil. European
Journal of Soil Science 58, 763774.
Medvedev, V.V., Cybulko, W.G., 1995. Soil criteria for assessing the maximum
permissible ground pressure of agricultural vehicles on Chernozem soils. Soil
& Tillage Research 36, 153164.
Pebesma, E.J., 2004. Multivariable geostatistics in S: the gstat package. Computers
and Geosciences 30, 683691.
R Development Core Team, 2009. R: A Language and Environment for Statistical
Computing. R Foundation for Statistical Computing, Vienna, p. 409.
Schunn, C.D., Wallach, D., 2005. Evaluating goodness-of-t in comparison of models
to data. Psychologie der Kognition: Reden and Vortrage anlasslich der Emeritierung von Werner Tack, 115154.
Soane, B.D., van Ouwerkerk, C., 1995. Implications of soil compaction in crop
production for the quality of the environment. Soil & Tillage Research 35, 522.
Webster, R., Oliver, M.A., 2007. Geostatistics for Environmental Scientists (Statistics
in Practice). Wiley, San Francisco, p. 286.