Content uploaded by Stéphane Sammartino
Author content
All content in this area was uploaded by Stéphane Sammartino on Oct 13, 2015
Content may be subject to copyright.
Vadose Zone Journal | Advancing Critical Zone Science
Identifying the Functional
Macropore Network Related
to Preferential Flow in
Structured Soils
Stéphane Sammartino,* Anne-Sophie Lissy, Christina
Bogner, Romain Van Den Bogaert, Yvan Capowiez,
Stéphane Ruy, and Sophie Cornu
Understanding the processes and mechanisms that control preferential flow
in soils in relation to the properties of their structures is still challenging since
fast flow and transport occur in a small fraction of the porosity, that is, the
functional macropore network, making it difficult to image and character-
ize these processes at decimeter scales. The aim of the paper was therefore
to propose a new image acquisition and analysis methodology to charac-
terize preferential flow at the core scale and identify the resulting active
macropore network. Water infiltration was monitored by a sequence of
three-dimensional images (taken at 5 -, 10-, or 15-min inter vals) with an X-ray
scanner that allows very fast acquisitions (10 s for a 135-mm diameter). A
simultaneous dye tracer experiment was also conducted. Water infiltration
was then imaged at each acquisition time by the voxels impacted by water
during infiltration, named the water voxels. The number of times a voxel was
impacted by water during the experiment was converted into data reflect-
ing the water detection frequency at the given position in the soil column,
named the local detection frequency. Compared with dye staining, the
active macropore network was defined by macropores in which water
voxels were the most frequently detected during the experiment (local
detection frequency above 65%). The geometric proper ties of this active
network, such as the connectivity, were significantly different from those of
the total structure. This image processing methodology coupled to dynamic
acquisitions can be used to improve the analysis of preferential flow pro-
cesses related to soil structures at the core scale.
Abbreviations: BB, Brilliant Blue FCF; EPC, Euler–Poincaré characteristic; HU, Hounsfield
unit; IMC, integral of mean curvature; PSD, pore size distribution; X- CT, X-ray computed
tomography.
The role of macropores in fast f low and transport phenomena occurring in soils
has long been recognized, and during the last 30 years, interest sharply increased because
of the attention paid to environmental issues (Beven and Germann, 1982, 2013). Although
water transfer and solute transport in soils is still largely modeled by Darcy–Richards and
advection–dispersion equations, non-equilibrium models such as dual permeability (Gerke,
2006) have been developed to account for fast flow and transport in macropores (Jarvis
et al., 1991; Di Pietro et al., 2003; Germann et al., 2007; Simunek et al., 2008; Nimmo,
2010). While these numerical approaches can help to reproduce hydrographs and break
-
through curves, they still suffer from a lack of generality and lack of explicit links with
soil structure. The weaknesses of these models lie in the definition of the mass exchange
parameters that describe processes and mechanisms at the interface between macropores
and the soil matrix, and in the assessment of the real effective exchange surface involved
in a given experimental context (Gerke, 2012). Up to now, such parameters were often
empirically defined or were only parameters obtained by fitting that were used to match
numerical and experimental results.
We proposed a method to identify
the functional part of the macropo-
rosity of structured soils in relation to
preferential flow pathways based
on (i) recording water infiltration by
a sequence of three-dimensional
images with an X-ray scanner and
(ii) a new data processing reflect-
ing the local detection frequency
of water voxels in macropores. The
goal is to better define the func-
tional part of a structure when
preferential flow occurs and to gain
a greater insight into what controls
fast flow and transport phenomena
in structured soils.
S. Sammartino, UAPV, UMR 1114
“Environnement Méditerranéen et
Modél isation des Agro-Hydrosystèmes,”
Site Agroparc, F- 84914 Avignon Cedex 09,
France; A. -S. Lis sy, R. Van Den Bogaer t, and
S. Ruy, INRA, UMR 1114 “Environnement
Méditerranéen et Modélisation des Agro-
Hydrosystèmes,” Site Agroparc, F-84914
Avignon Cedex 09, France; C. Bogner,
Ecolog ical Modell ing, BayCEER, Univ.
of Bayreuth, Dr.-Hans-Fr isch -Straße 1-3,
95448 Bayreuth, Germany; R. Van Den
Bogaert and S. Cornu, INRA, UR 1119
“Géochimie des Sols et des Eaux ,” F-13100
Aix en Provence Cedex 04, France; Y.
Capowiez, INRA, UR 1115 “Plantes et
Systèmes de Culture Hor ticoles,” Site
Agroparc, F-84914 Avignon Cedex
09, France. *Cor responding author
(stephane.sammartino@univ-avignon.fr).
Vadose Zone J.
doi:10.2136/vzj2015.05.0070
Received 7 May 2015.
Accepted 21 July 2015.
Original Research
© Soil Science Society of Amer ica
5585 Guil ford Rd., Madison, WI 53711 USA.
All rights re ser ved.
Published October 12, 2015
VZJ | Advancing Critical Zone Science p. 2 of 16
Recently, Gerke (2012) proposed a new method to assess the inter-
domain surface in structured soils on the basis of the processing
of X-CT images. Although this analysis proved unable to quantif y
a well-defined interface it restricted the range of possible values.
This approach seems promising because the surface is obtained
through measurements and not from data fitting or inverse model-
ing. However, it does not account for the occurrence of preferential
flow, which is ubiquitous in structured soils (Nimmo, 2012). As
pointed out by Beven and Germann (2013), there are still basic
unanswered questions about water flow in macropores, such as
“How does water flow through macropores in the soil?” and “How
does water in macropores interact with water in the surrounding
soil”? These questions can now be approached with the help of
internal imaging techniques for visualizing f low in soil structure
during experiments performed under unsaturated conditions. New
developments in image acquisition and analysis are also needed
(Stampanoni et al., 2006; Sammartino et al., 2012; Koestel and
Larsbo, 2014).
Although dye tracing has been widely used at the profile or core
scales to identify flow pathways and flow regimes in soils (Mooney
and Morris, 2008; Kodesova et al., 2012; Kasteel et al., 2013), this
technique is destructive and cannot trace the dynamics of water
flow within a rainfall event. Nowadays, X-CT is more exten-
sively used as a nondestructive way to image the inner structure
of soils, combining mineral composition, water content, and
bulk density data (Mees, 2003; Kalender, 2005; Helliwell et al.,
2013; Wildenschild and Sheppard, 2013). The three-dimensional
mapping of the attenuation coefficients obtained in natural
media is well adapted to detect water, air, and solids (Ketcham
and Carlson, 2001; Mees, 2003; Kalender, 2005). To date, how-
ever, X-CT was mainly dedicated to the characterization of soil
structure for comparison with hydrodynamic properties or to
monitor structure evolution under various external factors (Luo
et al., 2010; Capowiez et al., 2014; Naveed et al., 2014; Katuwal et
al., 2015). Most of the studies imaging the spatial distributions of
water or tracers (contrast agents) in soil structure were essentially
conducted either at hydraulic equilibrium or at saturation under
stationary flow (Mooney, 2002; Luo et al., 2008; Luo and Lin,
2009; Tippkötter et al., 2009). A technical hindrance to achiev-
ing dynamic experiments is the acquisition duration when using
industrial X-CT, which is rarely less than half an hour. However,
recently, dynamic recordings were undertaken at the centimeter
scale (a few centimeters to less than 1 cm) using a new acquisi-
tion mode with a constant rotation instead of a sequential one
to shorten scan durations to a few minutes in industrial X-CT
(Flavel et al., 2012; Helliwell et al., 2013; Koestel and Larsbo,
2014). Some dynamic experimental studies have also been designed
in synchrotrons to benefit from higher photon fluxes. Scan dura-
tions of the order of a few to 1 s were obtained, but samples were
even smaller because of the smaller fields of view (Youssef et al.,
2010; Berg et al., 2014; Leu et al., 2014). With respect to these
new technical advances, two points may be highlighted: (i) the
centimetric sample sizes used are not adequate to account for soil
structure effects on mass transfer processes and soil structure is a
key controlling factor (Vogel et al., 2006). Therefore, experiments
have to be performed at a decimetric or larger scale; (ii) as soil
macroporosity is generally unsaturated, experiments conducted
in saturated conditions poorly characterize the complex flow pro-
cesses and flow regimes (Ghezzehei and Or, 2005; Jarvis, 2007;
Nimmo, 2010; Vogel et al., 2010) encountered in unsaturated con-
ditions. Besides, Clothier et al. (2008) and Nimmo (2012) also
showed that many, if not most, natural soils exhibit at one time or
another a degree of hydrophobicity that has the effect of trigger-
ing preferential infiltration well below saturation. Therefore, we
recently designed a new visualization method to characterize f low
dynamics in unsaturated and decimetric soil cores on the basis of
time-resolved acquisitions made during simulated rainfall events
in a medical scanner (Sammartino et al., 2012). The advantage of
the latest generation of medical scanners is that they combine the
helical acquisition mode with the multislice recording, thus reduc-
ing the scan duration of soil cores to a few seconds. The processing
of these dynamic data is based on a new approach that accounts
for the local detection frequency of the water voxels in macropores,
that is, considering the number of times a given voxel was impacted
by water over the experiment.
In this study, we propose a new image acquisition and analysis
approach to characterize preferential f low at the core scale and
identify the resulting active macropore network. This recent
development in X-CT images was compared with the staining of
preferential flow pathways to assess its efficiency at characterizing
fast flow and transport phenomena in unsaturated macroporous
soils. The objective was to test the applicability of the methodology
for the identification of the functional part of the soil structure.
This active structure corresponds to the macropore network in
which preferential flow and mass exchanges with the soil matrix
occur. The geometrical properties of the active macropore network
were compared with those of the entire structure as well as the
distribution of the water voxels as a function of their detection
frequency over the experiment.
6Material and Methods
Soil Samples
The two soil cores, referenced L20-1 and L6-4, were initially
sampled from 35- to 50-cm depth in a cultivated field near Saint-
Ouen-de-la-Cour, France (48.407° N lat, 0.590° E long) to study
clay translocation in an eluvial horizon of Luvisol (Cornu et al.,
2014). To collect the samples, PVC tubes were slowly and vertically
pressed into the soil. The diameter and height of the soil cores
were 135 and 150 mm, respectively. The sampled horizon was
characterized by a yellow-brown color and a clay loam texture
(Table 1). The cores underwent a succession of 30 rainfall events
of 30 mm performed with a rain intensity of either 20 mm h−1 or
VZJ | Advancing Critical Zone Science p. 3 of 16
6 mm h
−1
for L20-1 and L6- 4, respectively (see Cornu et al. [2014]
for more details). These experiments showed that the hydraulic
properties in these two columns probably differed despite their
being sampled in the same soil horizon.
X-ray Tomography
The Siemens “SOMATOM Definition AS” X-CT located in
the “Val de Loire” center of the French National Institute for
Agricultural Research (INRA) was used to acquire the sequences
of time-resolved images. This latest generation of multislice helical
scanners includes 64 multidetector rows. It thus allows very fast
acquisitions by recording a maximum of 128 slices simultaneously
with an image quality fully comparable to that of a single-slice
scanner (Klingenbeck-Regn et al., 1999). After preliminary tests,
the following settings were chosen and applied to all scans: tube
voltage of 140 kV, tube current of 175 mA, and pitch factor of
0.35. The pitch factor corresponds to the table travel per rotation
divided by the effective detector row thickness. A pitch factor
lower than one reduces artifacts and noise by overlapping acqui-
sitions. However, multidetector rows work efficiently at certain
pitch values, and the recommended values were followed. More
technical details about medical scanners and their setting param-
eters can be found for example in Kalender (2005) and Goldman
(2008). The speed of the scanner table was set at 13.4 mm s
−1
. The
entire soil volume was thus scanned in 10 s. With a field of view
of 170 mm in diameter and a 512 by 512 numerical matrix, the
pixel size (resolution in slices) was 332 µm. The slice thickness
was set at its minimum size, that is, 600 µm. Tomographic images
were reconstructed with the Siemens internal software using a
filter enhancing the object edges and an automatic correction of
beam hardening. Therefore, no beam hardening was present in the
reconstructed tomographic sections. Sections were automatically
calibrated in density according to the Hounsfield unit scale (HU)
(Hounsfield, 1980) and exported in the standard Digital Imaging
and Communications in Medicine (DICOM) format encoded in
signed 16-bit.
Time-Resolved Image Sequence and Dye
Tracing Experiment inside the Scanner
The L20-1 and L6-4 columns were submitted respectively to a rain
intensity of 20 mm h
−1
, chosen to mimic a summer storm event,
and of 6 mm h−1, to represent a typical winter rain event. Before
the start of the simulated rain, they were close to field capacity, that
is, the soil has been saturated and allowed to drain freely for more
than 48 h. The simulated rainfal l events were conducted inside the
scanner with the help of a specific homemade device. The wood
structure that supported the columns avoided a strong attenua-
tion of X-rays. Appropriate rain intensities and durations were
applied using a rainfall simulator connected to a pulse pump. More
detailed technical information can be found in Sammartino et al.
(2012). In the present study, the effluent was collected at the outlet
of the column but was not usable quantitatively because of experi-
mental problems. The column placed vertically on the scanner
table was subjected to the rainfall with a dye added to the rainwa-
ter. During the simulated rainfall event, several three-dimensional
images were acquired from the start to the end of the infiltration
experiment as reported in Table 2. Before the start of the simulated
rain, a three-dimensional image (reference stack) was scanned to
record the initial distribution of water within the column. During
the experiment, the water infiltration was monitored while the dye
transported by rainwater diffused through macropore surfaces and
was absorbed around the active macropores.
The food dye Brilliant Blue FCF (BB) classically used in soils for
the staining of flow pathways was chosen (Bogner et al., 2008;
Anderson et al., 2009). Its sorption and desorption characteristics
were studied by Morris et al. (2008) as a function of clay content.
Its properties were considered appropriate to mark the flow path-
ways in the studied soil. The molecule (C37H34N2Na2O9S3)
with a molar mass of 792.9 g mol−1 is stable for a large range of
pH and ionic strengths (Flury and Fluhler, 1994; Persson et al.,
2005; Kasteel et al., 2007). Although tracing with BB is a classi-
cal experiment in soils, the color and the visibility of the resulting
flow patterns depend strongly on the hue, the structure, and the
clay content of soil samples. The appropriate concentration of the
dye solution has therefore to be determined before the rainfall
Table 2. Characteristics of the simulated rainfall events and of time-
resolved image sequences.
Soil
core
Rainfa ll
intensity
Rainfa ll
duration
Number
of scans
during
rainfall
Time-resolved i mage sequences
(Time af ter the start of the rai nfall
application)†
mm h−1 min min
L2 0 -1 20 90 13 2, 7, 12, 17, 22, 27, 32, 42, 52 , 62, 72,
82, 92
L6-4 6300 21 2, 17, 32, 47, 62, 77, 92, 107, 122, 137,
159, 167, 182, 197, 212, 230 , 245,
260, 275, 290, 302
† Recording times in b old were used for preferential f low path recognition.
Table 1. Characteristics of the sampled horizon after Cornu et al.
(2014).
Characteristics S ampled horizon
Sampling depth, m 0.35–0.50
Clay fract ion, kg kg−1 0.307
Fine silt fra ction, kg kg−1 0.285
Coarse si lt fraction, kg kg−1 0.302
Fine sand fraction, kg k g−1 0.080
Coarse sa nd fraction, kg kg−1 0.026
Organic carbon, kg kg−1 0.0035
Bulk density, kg m−3 1.6
Porosity, m3 m−3 0.40
Water content at saturation, m3 m−3 0.37
VZJ | Advancing Critical Zone Science p. 4 of 16
simulations. After several tests on columns of the same soil (dye
concentration, application duration, and lag time before cutting),
the dye concentration was set at 4 g L
−1
and applied throughout
the rainfall duration. The time between the end of the rainfall
and the beginning of the core cutting was 24 h. With the chosen
experimental conditions, the dye was still in excess in the drain-
age water and the blue patches were saturated even in the bottom
cross-sections. The effect of dye sorption was thus shown to be
negligible, ensuring a good staining of the preferential flow paths
on the entire height of the column.
Acquisition and Processing of Color
Images of Dye Distribution
A specific device was designed to allow accurate cutting of the
column perpendicular to its axis with a constant and controlled
cutting thick ness (Fig. 1). The soil was pushed out of the PVC tube
using a hydraulic jack and cut every 5 ± 1 mm with a well-sharp-
ened thin spatula. As the soil volume moved quite easily inside the
PVC tube when the hydraulic jack was activated, it can be assumed
that soil cores were not submitted to strong mechanical stress. The
new setup also maintained the camera and the illumination device
(four tungsten-filament lights of 150 W each) at a constant dis-
tance from the soil surface. Cutting and imaging of the stained
sections were performed in accordance with the recommendations
set out in Weiler and Fluhler (2004), Persson et al. (2005) and
Mooney and Morris (2008): the cutting method, removal of dust
before taking pictures, correction of geometric bias, color calibra-
tion, etc. Then, a gray scale and a color control patch (Kodak) were
added to all the pictures that were stored in the RGB RAW format
with a spatial resolution of 82 microns. A calibration plate was also
used to correct the geometric distortion.
For the purpose of the study, binarization of the cross-section
images in stained and unstained areas was chosen instead of
calibration of the intensity of the stained patterns against the con-
centration of BB, since binarization gives rapid access to the spatial
distribution of the dye in a section (Bogner et al., 2013). The seg-
mentation of a blue hue on a brown background was improved by
converting the RGB color photographs into HSV, as proposed by
Weiler and Fluhler (2004). Image processing of the stained sections
followed four main steps: (i) correction of the white balance; (ii)
correction of the geometric distortion and color normalization;
(iii) conversion to HSV color space; and (iv) binarization by an
automatic thresholding depending on the soil hue of each section.
An ad hoc procedure described in Bogner et al. (2013) was used for
the last step. After thresholding blue patches in the cross-sections,
the outlines of the binary objects were extracted and then “super-
imposed” on the color images in a transparent paste mode. The
quality of this thresholding was checked by comparing the object
outlines with those of the blue patches and by verifying that none
were missing or in excess.
Processing of X-CT Image Sequences
The processing of time-resolved image sequences adapted from
Sammartino et al. (2012) is summarized in Fig. 2. In the first
steps, 1 and 2, the raw data were formatted. Then, all the three-
dimensional images were resized with a cubic spatial resolution of
332 µm. In the following steps, 3 and 4, two data sets were acquired,
namely soil structure by thresholding the macroporosity and the
spatial distribution of water infiltration by thresholding the water
voxels at each recording time of the image sequence. As proposed
before, owing to the coarse size of voxels and to the effect of partial
volume averaging (Barrett and Keat, 2004), the soil attenuation
was considered as a linear function of the attenuations of water, air,
and soil matrix, respectively (Fig. 2, part 3), and the thresholdings
were made with assumptions on the possible combinations of these
three main phases.
The macroporosity was defined by (i) considering the macropores,
that is, pores larger than the resolution, by all possible combina-
tions of air and water in Hounsfield density, and (ii) considering
part of the macropores under the resolution, called the diffuse
macroporosity. Indeed, the voxel size of 332 µm appears to be
limited to define the macroporosity. We hypothesize that if a
voxel is composed half of soil matrix and half of water or air, it
is likely to belong to a macropore. We used the HU density to
Fig. 1. Setup for the cutting of soil columns and the recording of pictures.
VZJ | Advancing Critical Zone Science p. 5 of 16
calculate the threshold depending on
the peak of the soil matrix. The diffuse
macroporosity thus refers to the voxels
containing a volume filled to at least
50% by air or water, and the threshold
between soil matrix and macroporosity
was set to half of the peak of the soil
matrix (Fig. 2, part 3).
The thresholding of the water voxels
was based on difference images as
we were only interested in the water
in motion in the core owing to
infiltration and not in all the water
content. Consequently, we needed
to record the initial distribution of
water and air—stack (t0) in part 4 of
Fig. 2—before applying the rainfall to
reveal the water–air motions within
the sample by the stack difference (t) at
each time recording t. The water voxels
are thus the voxels whose HU density
was modified by the infiltrating water
at each time step compared with the
initial water distribution before the
start of the rainfall. To threshold
the water voxels from the difference
stacks, we considered a range of
Hounsfield densities around the mode
of the water-filled voxels: 1000 HU the
value of air voxels at t0 entirely filled
with the infiltration water at a time
t during the experiment. The same
type of assumptions as those made
for the diffuse macroporosity were
applied to set the thresholds LTw and
HTw around 1000 HU (Fig. 2, part
4). Consequently, the resulting water
voxels are not strictly the voxels totally
filled by the infiltrating water, but the
voxels in which at least 50% of the
volume was impacted by water during
the flow experiment. The threshold
values are reported in Table 3.
Data Processing
Frequency of Occurrence of Water Voxels
The number of times the voxels were categorized as water voxels
throughout the experiment was cumulated and recorded in a three-
dimensional grayscale image. A gray level of 0 means that the voxel
was never a water voxel during the experiment. A gray level of p for
the voxel (x, y, z) means that this voxel was impacted by water p
times over the n acquisitions, that is, with a detection frequency
of p/n at the position (x, y, z), named a local detection frequency.
Comparison of the Spatial Distributions of
Water Voxels and Brilliant Blue FCF
To compare the part of the structure seen by water with the part
of the structure marked by BB, an ad hoc method was used based
on a three-dimensional geodesic reconstruction of the subsets of
Fig. 2. Graphical abstract of image processing applied to X-ray computed tomography image sequences.
Pm, peak of the soil matrix in Hounsfield units (HU); ROI, region of interest. Full air or water voxels are
the voxels containing only air or water before the experiment, −1000 and 0 HU, respectively. Air-filled
or water-filled voxels are the voxels that have been completely drained or filled during the experiment,
−1000 or 1000 HU in difference stack. Voxels with HU ranging from LTw to HTw are considered as
being impacted by infiltration water during the recording time and called water voxels.
VZJ | Advancing Critical Zone Science p. 6 of 16
the structure from markers (either a set of water voxels or the BB
patches) in a mask (the entire structure). In mathematical mor-
phology, the geodesic reconstruction from markers in a mask is
classically based on the iteration of a geodesic dilation applied to
the markers in the mask up to the entire reconstruction of the
objects containing markers (Lantuejoul and Beucher, 1981; Serra,
1982). The various sets of water voxels were considered either at a
given recording time or in a range of the local detection frequencies.
For practical reasons, the BB distribution was reconstructed on the
column height by a bi-cubic interpolation from the cut sections.
The reconstructed stack was then resampled and rescaled to be
identical to the tomographic stack.
Before the geodesic reconstructions, the smallest markers were
removed using a 26-neighbor median filter. Then, the three-
dimensional structures reconstructed from the various sets of water
voxels chosen as “flow markers” were compared with the three-
dimensional structure reconstructed from dye patches for each
column. This comparison was made using geometric descriptors
of soil structure.
Geometric Descriptors of Soil Structure
The morphology and connectivity of the pore space and the pore
size distribution (PSD) are the fundamental characteristics of
soil structure that control soil properties (Vogel and Kretzschmar,
1996). As proposed by Vogel et al. (2010), a limited set of geometric
descriptors, named the Minkowski functionals, can be used to
characterize the complex structures of soils. These geometric
descriptors are (i) the pore space volume (V); (ii) the pore surface
area (S), which represents the interface between pores and solids
(soil matrix); (iii) the integral of mean curvature of this interface
(IMC), which characterizes the geometrical arrangement, that is,
the shape of the pore space; and (vi) a connectivity index, known
as the Euler–Poincaré characteristic (EPC) or Euler number,
which depends on the integral of total curvature of the pore–solid
interface.
Intuitively, the “curvature” is the amount by which a geometric
object deviates from being “f lat” sensu stricto or in average. Indeed,
the mean curvature at a point is calculated by the average of the
two principal curvatures (Levitz, 2007). The value is positive for a
convex surface and negative for a concave one. For convex objects,
the IMC is equivalent to a mean apparent diameter. For very elon-
gated objects (needles, fibers, macropores), the mean curvature is
the object length.
The EPC is calculated by the number of isolated clusters of objects
minus the number of redundant connections (redundant loops),
that is, the number of bonds that can be cut without creating
another isolated cluster. Thus, for a porous structure composed
of isolated structural pores, the EPC will be positive and tend
to be equal to the pore number. A complex well-interconnected
structure will furnish negative values of EPC.
These global geometric measurements do not provide information
on the PSD. We therefore used the classical PSD analysis by mor-
phological openings with a structuring element of increasing size
(Serra, 1982; Soille, 2004). Briefly, an opening (erosion followed
by a dilation) by a sphere of radius r removed the pores or throats
of pores smaller than 2r in any direction. Then, the complement
of the intersection with the initial set of porous objects gave the
required pore fraction with sizes less than 2r (quantifications com-
puted with the Software Avizo Fire 8.0.1). Each reconstructed
macropore network (as described in the section “Comparison
of the Spatial Distributions of Water Voxels and Brilliant Blue
FCF”) was characterized by this set of geometric descriptors. The
macroporous network reconstructed from dyed patches was con-
sidered as the reference functional structure of each soil sample
and named X_BB (X for L20-1 or L6-1). The comparison with the
macropore networks reconstructed from the water voxel sets was
made through the relative variations in their geometric descriptors.
The smallest relative variations define the best match between the
reference structure and the structures defined by voxels impacted
by water. A positive deviation overestimates the parameter com-
pared with that of the reference structure.
6Results
Qualitative Overview of the Data as a
Function of the Core Depth
X-CT Results: Relation between Macroporosity
and Water Voxels
The mean macroporosity values of core L20-1 and L6-4 were
4.9 and 7.8%, respectively. These values are of the same order of
magnitude as those commonly found in the literature (e.g., in
Katuwal et al., 2015). Thus, they account for only 10 to 20% of
the porosity of the soil horizon given in Table 2. The macroporos-
ity decreased with depth with a “surface layer” (0–4 cm) that was
more porous for both samples (Fig. 3a and 3c). The mean macro-
porosity values of that layer in core L20-1 and L6-4 were 8.4 and
18.0%, respectively. Below this surface layer, the macroporosity
ranged from 2% to 7 to 9% with a mean value of 4% for both
columns. Macroporosity peaks were observed around 7-cm and
Table 3. Thresholds used to define the soil macroporosity from the ref-
erence stack and the water voxels from the series of difference stack.
Soil core
Peak of soil
matrix
Threshold s for
macroporosity
Threshold s for water
voxels
——————————————— H U † ———————————————
L2 0 -1 1076 −1024 to 538 500 to 1538
L6-4 993 −1024 to 497 500 to 1497
† Hounsfield units.
VZJ | Advancing Critical Zone Science p. 7 of 16
12- to 13-cm depths in the L20-1 core. They were associated to
planar horizontal cracks.
On the cross-sections (Fig. 3b and 3d), it can be seen that the
voxels impacted by water (i) are distributed mainly inside
the macroporosity and (ii) are located either at the surface of
unsaturated macropores or completely fill a macropore. In addition,
(iii) some macropores do not contain any water voxels and (iv) some
water voxels are not located in the macroporosity but distributed
in the soil matrix (displayed in red on the cross-sections). These
voxels can be found throughout the column.
The three-dimensional image of the water voxels at five different
times after the start of the rainfall (Fig. 4) and the depth profile
distribution of these voxels (Fig. 3a and 3c) demonstrate the fol-
lowing: (i) water filling and storage in the surface layer and (ii)
water infiltration from the surface layer in some macropores that
progressively fill up. The last profiles acquired in the experiment
(at time 52 and 72 min for L20-1 and 212 and 292 min for L6-4)
were almost identical, suggesting that a “stationary state” was
reached in columns whatever the intensity considered. In addi-
tion, the depth distribution of macroporosity and the volume of
voxels impacted by water at the end of the experiment were similar
(R2 of 0.90 for L20-1 and 0.98 for L6-4), the latter remaining con-
siderably lower than the macroporosity. This suggests that a large
fraction of the macroporosity remained air filled.
Dye Results: Relation between X-CT
Macroporosity and Dye Distribution
We first observed from cross-sections that stained patches were
properly thresholded and that artifact macropores due to the
section cutting (e.g., Fig. 5c, center cross-section) were not thresh-
olded. After the dye experiment, both core samples were stained, by
the brilliant blue, over the whole column height. The core sections
were almost entirely dyed over the first 2 cm. Then, the evolutions
of macroporosity and dye coverage depth patterns are similar. The
dyed area decreased strongly in sample L20-1 to reach a mean value
close to 3%, with two peaks at more than 10% of dye coverage at
7- and 12-cm depths. They are related to the planar horizontal
cracks already mentioned. In sample L6-4, dye coverage decreased
slightly throughout the column, from 30 to 3%, with two less
marked peaks related to the macroporosity evolution. Very low dye
coverages are due to sparse stained patches of various sizes obser ved
on cross-sections. This suggests that the flow is strongly localized.
Fig. 3. Profiles of macroporosity (in black, m3 m−3) and content in water voxels for different times after the start of the rain (in various hues of blue, m3 m−3)
for samples L20-1 (a) and L6-4 (c). Horizontal cross-sections of the soil columns with the macroporosity (in black), the matrix (in orange), and all the
water voxels cumulated during the rain (in blue in macroporosity and in red in soil matrix), at 0.5-, 8.0-, and 11.0-cm depths, and for column L20-1 (b)
and column L6-4 (d). Soil core diameter is 135 mm.
VZJ | Advancing Critical Zone Science p. 8 of 16
Fig. 4. Three-dimensional representation of the water voxels at five different times after the start of the rainfall application (obtained by Isosurface
interpolation from AvizoFire 8.0.1 software). (a) Sample L6-4. (b) Sample L20-1.
Fig. 5. (a) Macroporosity and dye coverage
profiles of the two soil columns. Stained
sections showing Brilliant Blue FCF dye
distribution and its thresholding, at depths of
0.5, 8.0, 11.0 cm, for (b) column L20-1 and
(c) L6-4. The thresholding is shown by the
blue lines.
VZJ | Advancing Critical Zone Science p. 9 of 16
Overall, the dye coverage depth pattern of L6-4 was quite well
correlated with the macroporosity evolution, with a R
2
of 0.77. For
L20-1, the dye spreading in the planar horizontal cracks precluded
correlation between the profiles. The upper 4 cm of the columns
will not be considered in the following as they did not permit a
proper comparison between the active networks marked by dye
and the voxels impacted by water.
Comparison of Functional Structures
Obtained from the X-CT and Dye Images
Dening the Sets of Water Voxels as Potential
Flow Markers
As discussed on Fig. 3, the voxels impacted by water were not
strictly located in the macroporosity and might prove to be
artifacts. The question was thus to determine which voxels thres-
holded as being impacted by water ref lected preferential f low paths.
Considering the number of times each
water voxel was detected during the
experiment, three frequency domains
were defined: voxels poorly impacted
by water, with a rate less than 30%
during the experiment (“low” domain),
voxels impacted by water between 30
and 65% (“medium” domain), and
voxels strongly impacted by water,
with a rate more than 65% during
the experiment (“high” domain) (Fig.
6). In addition to the three groups of
frequencies, the “all” domain includes
all the voxels that were categorized as
water voxels at least once, and differ-
ent acquisition times were selected for
each column to be tested as potential
flow markers.
Comparing the Active
Macropore Networks
Obtained by the Two
Approaches
The three-dimensional geodesic recon-
structions obtained for the water voxels
at different acquisition times and for
the low detection frequencies, that is,
water voxels recorded with a rate less
than 30% at the same position during
the experiment, had a larger volume,
surface, and IMC than those obtained
from the BB staining, whatever the
column. These parameters were thus
considered as overestimated by the
X-CT approach. Conversely, their
connectivity (EPC) was lower and
Fig. 6. Histograms of the local detection frequency of water voxels
during the experiment.
Fig. 7. Relative variations of the geometric descriptors between the reference functional structure
reconstructed from the Brilliant Blue FCF approach and the structure reconstructed from the
different sets of water voxels. Numbers stand for the time at which the voxels impacted by water were
considered: “all” for the totality of the water voxels over the experiment and “low,” “medium,” and
“high” for the water voxels detected less than 30%, between 30 and 65%, and more than 65% on the
experiment duration, respectively. (a) Core L6-4. (b) Core L20-1. The rectangle display shows ± 10%
of relative variation over each axis since the axis extents vary. EPC, Euler–Poincaré characteristic;
IMC, integral of mean curvature.
VZJ | Advancing Critical Zone Science p. 10 of 16
thus underestimated (Fig. 7). The overestimations of the volume
and of the surface were usually lower than that of the IMC or
than the underestimation of the connectivity index. Whatever
the structural parameters, the largest overestimations and under-
estimations were always obtained when considering either all the
voxels impacted by water (“all” domain of local detection fre-
quency) or those seldom impacted by water (“low” domain), both
giving similar results. For the L6-4 sample, the comparison with
the acquisition times, 62 and 72 min after the start of the rain, also
gave acceptable results, with relative variations lower than 10%,
whereas none of the acquisition times of L20-1 gave a satisfac-
tory result. For that sample, the reconstructed structure with the
medium frequency domain (rate of detection ranging from 30 to
65%) also exhibited relative variations lower than 10%; this was
not the case, however, for L6-4.
The best match was obtained for both columns with the water
voxels in the high frequency domain, that is, the water voxels
recorded at the same position during the experiment with a rate
over 65% (Fig. 8). Note, however, that their residual structures,
that is, the difference between the two reconstructions, showed
some differences. For instance, in the case of L6-4, fine elongated
and vertical channels were not stained by the BB. These elongated
porosity features were not present in the residual structure of the
other sample. A more detailed morphological and topological
analysis needs to be done to better understand these differences,
but it is beyond the scope of this paper.
Finally, these results clearly indicate that the voxels most frequently
impacted by water during the experiment correctly trace
preferential flow paths marked by the dye tracer.
A Further Step in the Recognition of the
Active Macropore Network Related to
Preferential Flow Location
Comparison of Geometric Properties of Entire
and Active Structures
The active macropore networks comprised only 5% of the isolated
macropore clusters, but encompassed nearly 75% of the entire
macroporous volume and of the porous space surface (Table 4).
The IMC decreased drastically from the entire macroporosity to
the active macropore network. This characterizes more simple
macroporous surfaces due to earthworm burrows that mainly
represent the active macropore network, also associated with crack
planes for sample L20-1. As expected, active macropore networks
were more interconnected (negative EPC values) than the entire
structures (positive EPC values). Each active macropore network
included a percolating macropore involving nearly 85% of its
Fig. 8. Active structures defined by both Brilliant Blue FCF or by the water voxels the most often impacted by water during the experiment for L6-4
(a and b, respectively) and for L20–1 (d and e, respectively). (c and f) The complement of their intersection (XOR operator) for L6-4 and L20-1,
respectively. Volume reconstructions were displayed in Isosurface (Avizo Fire 8.0.1). The height of the bulk region of interest is 95 mm.
VZJ | Advancing Critical Zone Science p. 11 of 16
volume and clearly demonstrating the vertical continuity of the
connected structures (Fig. 9b and 9e).
All the geometric descriptors were significantly different between
the entire and active structures, the differences being larger for the
number of active isolated macropore clusters, the surface shape
(IMC), and the topology index (EPC).
The shapes of the PSDs of entire structures and active macropore
networks were very similar for both cores, with dominant pore
sizes in the range of 0.7 to 2 mm (Fig. 10). They were almost
identical for pore sizes larger than 2 mm. However, the fractions
of smaller pores (pore sizes below 2 mm) were lower in the active
macropore networks than in the entire structures.
Fig. 9. Three-dimensional representation of the entire structure (a and d), of active macropore networks (b and e) and the complement of their
intersection (XOR operator; c and f ) for L6-4 (a, b, and c) and L20-1 (d, e, and f ). Volume displayed using Isosurface (Avizo Fire 8.0.1). The considered
region of interest excludes the upper 0- to 4-cm surface layer. The active macropore networks of (b) and (e) are labeled in color considering one color
per three-dimensional object, that is, isolated macropore cluster.
Table 4. Geometrical descriptors of macroporosity and macropore networks, that is, active macropore network and percolating macropore.†
Soil core
Geometric al properties of entire
structure and of macrop ore networks
Number of isol ated
macropore clust ers EPC
Volume Surface IMC
10−5 % of A M % of ROI “bul k” 10−2 % of A M 103% of AM
m3m2m
L6-4 all macroporosity 6963 4922 5.4 10 0 4.1 13.6 100 2.6 100
active macropore network 128 −1831 4.7 87 3.5 10.5 77 1.3 48
percolating macropore 1 −179 7 4.0 74 3.0 8.7 64 0.9 35
L2 0 -1 al l macroporosity 6383 5538 4.7 100 3.5 13.8 100 2.6 10 0
active macropore network 277 −412 3.4 73 2.6 8.5 67 1.0 38
percolating macropore 1 −591 2 .9 62 2.2 7. 3 53 0.7 27
† ROI excluding the 0- to 4-cm upper layer of the columns. ROIs are the same size for the two cores. AM, all macroporosity; EPC, Euler–Poinca ré characteristic; IMC,
integra l of mean curvatu re; ROI, region of interest.
VZJ | Advancing Critical Zone Science p. 12 of 16
Distribution of Voxels Impacted by Water in
the Entire and Active Structures
The macroporosity and the water voxels were thresholded by
two independent methods. Most of the voxels impacted by water
during the experiment were distributed within the macroporosity,
but a fraction of them was also detected in the soil matrix (Table 5).
Nearly 80% of the water voxels located the macroporosity were also
in the active macropore network. This supports the assumption
that water flows preferentially inside the selected active structures.
Water voxels within the soil matrix were not uniformly distributed.
Most of them (64% for L6-4 and 89% for L20-1) were encountered
in the first millimeter surrounding the macroporosity.
When considering the voxels impacted by water in the low,
medium, and high frequency domains, very different localiza-
tions were obtained (Fig. 11). Voxels impacted with low frequencies
were more or less homogeneously distributed within the soil core
(matrix, macroporosity, and in the layer surrounding the macropo-
rosity). Theoretically, water voxels should not have been detected
in the soil matrix far from macropore edges as the soil matrix was
close to saturation before the start of the rainfall. Thus, the voxels
detected with low frequencies as water voxels and located in the
matrix indicate a noisy water voxel fraction of almost 10 to 20%.
The water voxels detected with a rate of more than 30% during the
experiment were mainly located in the macropores, with less than
2% of them in the matrix.
6Discussion
On the Staining of the Active Macropore
Network
The Brilliant Blue staining method was considered as a reference
method owing to the huge number of publications on this topic
that focused on the staining of preferential f low paths in soils. The
structure marked by the dye was thus considered as a “reference
functional structure” for the comparison with the distribution of
the water voxels. However, this is a cumbersome technique that can
also be prone to error under inappropriate experimental conditions.
The staining proved to be effective on the entire height of the
columns except in the upper 4 cm in which almost all the poros-
ity was saturated. This resulted from the diffusion of dye around
the active macropores that was greater than the average distance
between macropores. This upper 4 cm layer was not considered
further in the study as it was not possible to separate active and
inactive macropores. Under this layer, blue patches were scattered
and even in the lower cross-sections of the columns quite intensely
dyed. The high sorption of the dye in the upper part of the column
did not affect the staining of f low paths and their coloration. This
was also confirmed by the intense coloration of the water collected
at the outlet of the column after the breakthrough. Moreover, we
noticed that during the cutting of the core sections, other artifacts
may occur such as damage of the surface and spreading of dye by
the knife blade. However, they were not categorized as dyed areas
by our thresholding method and thus did not cause bias in the
resulting binary images.
Thresholding the color images in binary ones resulted in a
substantial loss of information since it can be assumed that the
Table 5. Distribution of water voxels in either the matrix, macroporos-
ity, or the active macropore net work.
Fraction of water voxel s L6-4 L2 0 -1
%
In the macroporosity 77.5 79.0
In the matrix 22.5 21.0
In the matrix surrounding t he macroporosity over a
thickness of 1 mm
14.4 18. 6
In the active network (normal ized to the macroporosity) 82.5 80.2
In the matrix surrounding t he macroporosity (normaliz ed
by the water voxel in the matrix)
88.7 63.7
Fig. 10. Pore size distribution for
the entire structures and the active
networks. (a) Core L6-4. (b) Core
L20-1. The labels of the x axis are the
upper limits of the class of size 0.66
mm in equivalent diameter. The pore
size distributions of active networks
were normalized to those of the entire
structure.
VZJ | Advancing Critical Zone Science p. 13 of 16
blue intensity will be correlated to the detection frequency of
water voxels in the active macropores. However, it is difficult to
establish a quantitative relationship between blue intensity and
dye concentration (Bogner et al., 2011), and this relationship is
not in any case necessary for the recognition of the active flow
paths. Thus, after binary and color images had been accurately
compared to validate the thresholding step, the dye experiment
and the processing of color images were considered to have been
properly done and were therefore used as a reference to validate the
tomographic approach.
On the Thresholding of the Water Voxels
The processing of large series of three-dimensional images requires
fast, robust, and reproducible techniques that are not observer
dependent. However, applying global thresholds on distributions
that are not highly bimodal can generate inaccuracies. Indeed,
it is likely that the low threshold chosen (LTw) may encompass
some noisy voxels in the set of water voxels (Fig. 2). Likewise, other
effects visible above the HTw threshold that should normally not
be accounted for in the present work cannot be totally excluded
from the set of water voxels. For instance, the values above HTw
may originate from small variations in the structure combined
with water infiltration due to local swelling of clay particles. On
the basis of the hypotheses made on voxel filling, however, this
should concern less than half of the voxel volume. Some other
minor changes may have occurred during rainfall such as changes
in the top surface of the column and/or the accumulation of
autochthonous particles within the structure. Lastly, the noise may
also evolve during the experiment owing to the increase in water
content, which increases the sample attenuation. Additionally,
the X-shaped artifact that results from the vertical position of the
column in the gantry also increases noise on the diagonals of the
central sections of the sample (Akin and Kovscek, 2003). As is
often the case in image analysis studies, the thresholding step and
noise are two critical points that can limit the extent to which the
method is fully quantitative. However, the fraction of noise voxels
lying in the matrix does not affect the results, and water voxels
lying in macropores were not used without being post-processed in
local detection frequencies. This post-processing can be considered
as a method to reduce noise and to improve the data by averaging
the difference images.
Comments on Core Scale versus Spatial
Resolution
The core scale (typically a cylinder with dimensions over 10 cm)
constrains the spatial resolution of X-CT images that is propor-
tional to the field of view and raises the problem of the definition
of the objects due to the partial volume artifact. Indeed, Lehmann
et al. (2006) demonstrated that the partial volume artifact was
negligible for voxel sizes 10 times lower than the studied objects,
and Vogel et al. (2010) systematically removed the objects with
an apparent diameter less than 5 voxels that were thus not well
defined. Although no absolute range for macropore sizes in soils
can be defined, their equivalent diameters are commonly in the
order of a few tens of microns (Rab et al., 2014) to a few millime-
ters. However when focusing on fast flow processes in soils, only
the macropores greater than 300 to 500 microns are the most
important (Jarvis, 2007; Luo et al., 2008). Even for the latter, the
spatial resolutions of medical scanners, ranging typically from 200
microns to a few millimeters, are still limited.
The ternary thresholding used makes it possible to partially over-
come the coarse resolution limitation (voxel size of 332 µm in this
study) by including a fraction of macropores with sizes less than
the resolution—the diffuse macroporosity—in the macroporosity
(Sammartino et al., 2012). The voxels categorized in the diffuse
macroporosity can have only 50% of their volume filled with air
(or water). This hypothesis states that in this case the voxel is likely
to belong to the macropores. The chosen value (50%) is however
heuristic and cannot be justified without knowing the PSD and
connectivity at the voxel scale. It constitutes a realistic working
hypothesis that improves the macroporosity connectivity at the
Fig. 11. Distribution of water voxels
detected for less than 30%, 30 to
65%, and more than 65% during the
experiment, that is, low, medium,
and high frequency domain in the
different soil compartments (matrix,
macroporosity, and the 1-mm layer
surrounding the macroporosity). (a)
Core L6-4. (b) Core L20-1.
VZJ | Advancing Critical Zone Science p. 14 of 16
core scale on the basis of the HU density. Recently, Scheibe et al.
(2015) followed the same thresholding approach. They showed
that without using this approach, the numerical simulations based
on the macroporosity obtained from X-CT images were not con-
sistent with the experimental results.
The calculation of the thresholds for the water voxels follows the
same methodology. The water voxels are defined for a water filling
of at least 50% of their volume. It thus allows for partial fillings
of voxels affected by water during infiltration. In theory, this
improves the detection of thick water films or rivulets of a few
tenths of a millimeter, but further studies are needed to quantify
this size limit of detection and its impact on the visualization of
free-surface flow processes at the macropore surface.
The Effect of Scan Duration
As stated by Wildenschild and Sheppard (2013), the acquisition
of tomographic images should be limited to situations where the
scanned object does not change during the scan time or when the
duration is considered to be “small” compared with the process
dynamics. The 215 wetting front velocities provided by Beven and
Germann (2013) were used to compare our scan duration with the
velocity of infiltration fronts in soils. The distribution of velocities
is centered on 0.3 mm s
−1
, and 80% of the given velocities are in
the range 0.1 to 1 mm s
−1
. The traveling distance of the wetting
front is thus much less than the spatial resolution for the acquisi
-
tion of one vertical section (0.044 s) and is in the order of 3 voxels
during the scan of the entire soil volume (10 s for 1 mm s−1 veloc-
ity). Thus, the newest multislice scanners can be exploited to image
fast flow processes without too many motion defects if the water
front velocities are less than 1 mm s−1.
Further Results on Macropore Flow
Except in the upper 4 cm of the cores, the flow distribution proved
to be highly preferential by the very low dye coverages of about
3% in sample L20-1 and about 10% in sample L6-4. Additionally,
more than 80% of the water voxels detected during the rainfall—
the voxels in which the infiltrating water was thresholded with
a detection limit of 0.0183 mm3 (half the voxel volume)—were
localized in the active macropore network, which represents less
than 10% of the soil porosity. Soil porosity was not characterized
in the present study since at field capacity water exchange between
the macroporosity and the matrix porosity is extremely limited and
thus the matrix porosity has a minimal impact on the given results.
The dye results are similar to those of Kasteel et al. (2013) obtained
on a Luvisol and a Cambisol. They showed that the first centime-
ters of their soil monoliths also acted as a redistribution volume
for the underlying macropores. The dye coverages were of the
same order of magnitude, showing the preferential distribution
of the flow. Whatever the rainfall intensity, the profile of water is
well correlated to the distribution of the macroporosity that how-
ever remains largely unsaturated similar to the results given in
Sammartino et al. (2012). These results suggest that close to field
capacity, preferential flow can also occur at a low rainfall intensity
of 6 mm h
−1
in the structure of the sample L6–4 mainly due to
its strong connection and vertical continuity. This can be related
to free-surface f low processes such as thick water films or rivulets
that commonly arise at the surface of large opened macropores in
soils. Very different flow patterns were, however, seen for the two
columns. We consider that they were due more to the differences
in the two structures than to the effect of rainfall intensity, but
the latter cannot be totally excluded.
The frequency of occurrence of water voxels in macropores was
helpful to recognize the active macroporosity compared with the
reference functional structure stained by the BB. Indeed, the use
of the water voxels recorded at a given acquisition time was not
efficient, possibly because the active macroporosity thus derived
was dynamic, whereas staining covers the entire experiment.
However, the result was not efficient with the water voxels detected
throughout the experiment either (“all ” frequency domain). In all
likelihood, this suggests that a fraction of the water voxels in mac-
ropores also corresponds to noisy and/or artifactual voxels as also
shown for the matrix. Thus, increasing the detection frequency
provides a better reconstruction of the active part of the macro-
porosity and helps to overcome the effects of noise and artifacts.
This is confirmed by the relative variations in the four geomet-
ric descriptors, which are less than 10% for the “high” frequency
domain.
The active macropore networks identified in our soil cores con-
sist of the most open and interconnected macropores that mainly
originate from earthworm activity. This macropore network may
also have been modified by the 30 rain–drying cycles applied to
the columns by Cornu et al. (2014), in particular in the E-horizon,
with possible clay migrations and modifications in the pore struc-
ture. However, we cannot compare the differences between the
two active macropore networks as a function of the rainfall inten-
sity because the “final” active structures that were characterized
strongly depend on the “initial” structures before the rainfall–
drying cycles.
The highly negative values of the EPC indicate the dominance of
redundant connections within the active network compared with
the number of isolated macropores. However, the EPC character-
izes the degree of interconnection of the identified structure but
does not ensure that this structure connects the upper and lower
surfaces of the studied soil volume. This is confirmed however
by the occurrence of a percolating cluster that shows the vertical
connection of each active macropore network. Besides, other kinds
of macroporous structures, such as the horizontal crack planes
present in sample L20-1, can significantly change the geometric
properties of the active structure and also be associated to other
hydraulic functions, such as horizontal drainage or water storage
(Cornu et al., 2014).
VZJ | Advancing Critical Zone Science p. 15 of 16
6Conclusions
The monitoring of water infiltration at the core scale with a
multislice medical scanner has made it possible to identity the
functional part of the macroporosity in a structured soil related
to preferential flow processes. The method used was validated by
a quantitative comparison with the f low pathways stained by a dye
tracer on the same core samples and during the same experiments.
The results were obtained in a realistic experimental context to
mimic the functioning of natural soil, that is, undisturbed soil
structure and unsaturated flow condition.
It was especially shown that the geometric properties of the active
macropore network, particularly those associated to the surface of
the pore space and the connectivity, differ strongly from those of
the total structure. Thus as suggested in recent papers (Capowiez
et al., 2014; Katuwal et al., 2015), the functional part of a struc-
ture should be more relevant than the total macroporosity usually
considered (i) in studies that attempt to identify the relationships
between structure and mass transfer properties and (ii) in flow and
transport models applied to natural soils.
The new image processing developed here, which (i) takes informa-
tion on structure and water infiltration under the spatial resolution
by partial fillings of voxels into account and (ii) processes the
spatiotemporal distribution of water voxels in a local detection
frequency, enables a better integration of pore scale processes at the
column scale. Besides, frequency data of water voxels distributed
in each porosity compartment (soil matrix, diffuse macroporosity,
and macropores) constitute a promising tool for a better under-
standing of processes and mechanisms controlling preferential flow
phenomena and mass exchanges at the macropore surface.
This opens new perspectives to link structure features of porous
media to their functional properties and to better investigate the
factors governing the occurrence of preferential flow in soils in
various experimental contexts in relation to their structure. The
use of a contrast agent for X-ray would make it possible to quantify
the exchange between macropore and micropore porosit y domains.
These results should be compared with numerical modeling
performed by coupling the Darcy–R ichards equation in the matri x
domain to the propagation of a kinematic dispersive wave in the
macroporous domain or as already suggested a coupled model with
free-surface flow (Di Pietro et al., 2003; Nimmo, 2010).
Acknowledgments
We are grateful to B. Jouaud, F. Tison, and F. Lecompte for their technical support. S.S., A.-
S.L., and S.C. acknowledge the French ANR-Blanc program for funding the AGRIPED ANR-
2010-605 project and the INRA-EA department for their financial support. The authors are
grateful to the ERDF (European Regional Development Fund), the Région Indre-et-Loire,
and bodies that have financially supported the use of the medical scanner. S.S., L.C.B., and
S.R. acknowledge the Centre for Franco-Bavarian University Cooperation for its support of
their cooperation project. The authors thank the three anonymous reviewers for their helpful
comments that have significantly improved the manuscript.
References
Akin, S., and A .R. Kovscek. 20 03. Co mputed tomography in p etroleum
engineering research. Geol. Soc. Spec. Publ. 215:23–3 8.
Anderson, A .E., M . Weile r, Y. Alila, and R.O. Hudson. 2009. Dye staining
and excavation of a lateral p referential flow net work . Hydro l. Ear th
Syst. Sci. 13(6):935 –944. doi:10.5194/hess-13-935-2009
Barrett, J., and N. Keat. 20 04. A rtifacts in CT: Recognition a nd avoid-
ance. Radiographics 24(6):1679–1691. doi:10 .1148/rg .24 60 4 5 06 5
Berg, S., R. Arm strong, H. Ot t, A. Geo rgiadis, S.A. K lapp, A. Schwi ng, R.
Neiteler, N. Br usse e, A. Mak urat, L. Leu, F. Enzman n, J.- O. Schwar tz,
M. Wolf, F. Khan, M . Kersten, S. Irvi ne, and M . Stampanoni. 2014.
Multiphase flow in porou s rock imaged under dynamic flow co n-
ditions with fast X- ray com puted micro-tomog raphy. Petrophysics
55:3 04–312.
Beven, K., and P. Germann. 1982. Macropores and wate r-flow in soi ls.
Water Resou r. Res. 18(5):1311–1325. do i:10.10 29/ WR018i 0 05 p01311
Beven, K., and P. Germann. 2013. Macropores and water flow in soils
revisited. Water Res our. Res. 49(6):3071–30 92. do i:10.100 2/w rcr. 20156
Bogne r, C., I. Schmied inge r, and B. Hu we. 2011. Rapid estim ation of Bri l-
liant B lue concentrations in soi ls by vis diff use reflectance spect ros-
copy. Geoderma 164:95 –98. doi:10.1016/j .geod er ma .2 011.04 .017
Bogne r, C., B. Tran con y Wideman n, and H. Lange. 2 013. Charac terising
flow patterns in soils by feature ex traction and multiple co nsensus
clustering. Ecol. Inform. 15:44 –52. do i:10.1016/ j.e coinf.2 013. 03 .0 01
Bogne r, C., B. Wolf, M. Schlather, and B. Huwe. 200 8. Ana lysi ng flow
patterns from dye tracer experi ments i n a forest soil using ex trem e
value statistics. E ur. J. Soil Sci. 59(1):103–113. d o i: 10 .1111 / j .13 6 5 -
2389.20 07.00974.x
Capowi ez, Y., S. Samm artino, and E. Michel. 2 014. Burrow systems of
endog eic ea rthworms: Effects of ear thwo rm abundan ce and con-
sequences for soil wate r infiltration. Pedobiolog ia 57(4– 6):30 3– 30 9.
doi :10.1016/j .p ed obi.2014. 04.001
Clothi er, B.E., S.R. Green, and M. Deu rer. 20 08. Preferential flow and
trans por t in soi l: Prog ress and prognosis. Eur. J. So il Sci. 59(1):2–13.
d oi :1 0. 1111/ j .13 6 5 -2 3 8 9. 2 0 07. 0 0 9 91. x
Cornu, S., L. Quenard, I. Cousin, and A. Samouelian. 2014. Experimental
approach of lessivage: Quantification and mechanisms. Geoderma
213(SI): 357–370.
Di Pietro, L., S. Ruy, and Y. Capowiez. 2 003. Pred icting prefe rential water
flow in s oils by travel ing -di spersive waves. J. Hyd rol. 278(1–4):64 –75.
doi :10.1016/S 002 2-1694( 03 )0 0124 - 0
Flavel, R.J., C.N. Guppy, M. Tig he, M. Watt, A. McNeill, a nd I. M. Young.
2012. Non- destructive qua ntification of cerea l roots in soil using
high -resolut ion X- ray tomograp hy. J. Exp. Bot. 6 3(7):25 03 –2511.
doi:10.109 3/j xb/err421
Flur y, M., and H . Fluh ler. 1994. Br ill iant blue FCF a dye trace r for sol -
ute transpor t studies —A toxicological overview. J. Envi ron. Qual.
23(5 ):110 8 –1112 . doi:10.2134/jeq1994.00472425002300050037x
Gerke, H. 2006. Prefe rential flow d escr iptio ns for struct ured so ils. J. Plant
Nutr. Soi l Sci. 169(3):382 –4 00. d oi:10.1002/ jp ln. 20 0521955
Gerke, H .H. 2012. Macros copic repres entation of th e interface b etween
flow do main s in structured soil. Vados e Zone J. 11(3). doi :10 . 2136 /
vz j2 011.0125
Germ ann, P., A. Helbling, and T. Vadi longa. 20 07. Rivulet ap proach
to rates of preferenti al inf iltration. Vadose Zone J. 6(2):207–2 20.
doi:10.2136/vzj2006.0115
Ghezzehei, T., and D. Or. 20 05. L iqui d fragmentati on and i nterm ittent
flow reg imes in unsaturated fractured media. Water Resour. Res.
41:W12406. doi:10.1029/2004WR003834
Goldman, L.W. 20 08. Prin cipl es of CT: Multisl ice CT. J. Nucl. Med. Tech-
nol. 36(2):57–68 . doi:10.2967/jnmt.107.04 4826
Hell iwel l, J.R., C.J. Stu rrock, K. M. Gra yling, S.R. Tracy, R.J. Flavel, I .M.
Young, W.R. Wha lley, and S.J. Mooney. 2013. Appl icati ons of X- ray
computed tomog raphy fo r exami ning biophysical interaction s
and structural development in soil sys tems: A rev iew. Eur. J. Soi l Sci.
64(3):279 –297. do i :1 0 .1111 /e j s s . 12 0 2 8
Hounsfie ld, G. 1980. Computed medi cal imagin g. Science
210(4 465): 22– 28 . doi:10.1126/science.6997993
Jar vis, N.J. 2007. A review of non -equil ibr ium water flow a nd sol ute
trans por t in soi l macropores: Principles, controlling fa ctors a nd
conseq uences for water q uality. Eur. J. So il Sci. 58(3):523 – 546.
d oi :1 0. 1111/ j .13 6 5 -2 3 8 9. 2 0 07. 0 0 91 5. x
VZJ | Advancing Critical Zone Science p. 16 of 16
Jar vis, N., P. Jansson, P. Dik, and I. Mes sing. 1991. Modeling water and
solute transpor t in mac roporous soi l.1. Model descri ption and sen -
sitiv ity analysis. J. S oil Sci. 42(1):59–70. d o i: 10 . 1111/ j .1 3 6 5 - 2 3 8 9.1 99 1.
tb00091.x
Kalender, W. 2005. Computed tomography—Fundamentals, system
techno logy, image qualit y, appli cations. 2nd revised ed. Publ icis
Corp. Pu bl., E rlangen, G ermany.
Kasteel, R., P. Garnie r, P. Vachier, and Y. Coquet. 2007. Dye tracer in -
filt ratio n in the p lough layer a fter straw incorporation. Geoder ma
137(3-4):360–369. doi:10.1016/j.geoderma.2006.08.033
Kasteel, R., F. Schnit zler, A.E. Be rns, J. Vander borght, and H. Veree cken.
2013. Visu alization of transpor t pathways for o rganic compounds in
undistur bed so il mon oliths. Geoder ma 195–196:70 –78. do i:10.1016/j.
ge od erm a. 2 012.11.014
Katuwal, S., T. Norgaard, P. Moldrup, M. La mande, D. Wild ensch ild, a nd
L.W. de Jonge. 2 015. Linking ai r and wate r transpor t in intact soi ls to
macropore characteristics inferred from X- ray computed to mogra-
phy. Geoderma 237–238:9–20. doi:10.1016/j.geoderma.2014.08.006
Ketcham, R., a nd W. Carl son. 2 001. Acquisition, opti mization and inter-
pretation of X-ray computed tomographic imager y: Ap plication s
to the geoscien ces. Com put. Geo sci. 27(4):381– 40 0. doi:10.1016/
S0098-3004(00)00116-3
Klingenb eck-Regn, K., S. Schal ler, T. Flohr, B. Ohnesorge, A. Kopp, and U.
Baum. 1999. Subsecond m ulti -sl ice com puted tom ography: Basics
and app lications. Eur. J. Rad iol. 31(2):110–124. doi:10.1016/S0720 -
048X(99)00086-8
Kodesova, R., K . Nemecek, V. Kodes, and A. Zig ova. 2012. Using dye
tracer fo r visuali zatio n of preferentia l flow at m acro - and mi -
croscales. Vadose Zone J. 11(1). doi:10.2136/vzj2011.0088
Koestel, J., and M. Larsb o. 2014. Imag ing and quantification of pref-
erenti al sol ute transport in soil ma cropo res. Water Resour. Res.
50(5):4357–4378. doi:10.1002/2014WR015351
Lantuejoul, C., and S. Beucher. 1981. On the us e of the ge odesic metr ic
in image- analysis. J. Microsc. 121:39–49.
Lehmann, P., P. Wyss, A. Flisch, E. Lehm ann, P. Vontobel, M. Krafcz yk, A .
Kaestner, F. Beckmann, A. Gygi, and H. Fluhl er. 20 06. Tomograp hi-
cal ima ging and mathematical descri ption of porous media used
for the p rediction of f luid d ist ribution . Vadose Zone J. 5(1):80–97.
doi:10.2136/vzj2004.0177
Leu, L., S. Be rg, F. Enzmann, R.T. Ar mst rong, and M. Ker sten. 2 014. Fast
X-ray m icro -tomography of multiphase flow i n berea sands tone:
A sensitivity stu dy on im age processing. Transp. Porous Media
105(2):451–4 69. doi :10.100 7/s11242 - 014 - 0378 - 4
Levitz, P. 2007. Toolbox for 3D i magi ng and m odel ing of porous m e-
dia: Relationship with transport properties. Cement Concr. Res.
37(3):351–359. doi:10.1016/j.cemconres.2006.08.0 04
Luo, L., and H. Lin. 200 9. Lacunar ity and fractal analyses of soil mac-
ropores and preferential transport using micro-X-ray computed
tomography. Vadose Zone J. 8(1):233–241. doi:10.2136/vz j200 8. 0010
Luo, L., H. Lin, and P. Hall eck. 2 00 8. Qua ntif ying soil structure an d prefer-
ential flow in intact soil usi ng X-ray computed tomo graphy. Soil Sci.
Soc. Am. J. 72(4):1058–1069. doi:10.2136/sssaj2007.0179
Luo, L., H. Lin, and J. Schmi dt. 2010. Qua ntitative relat ions hips betwee n
soil macropore characteristics and preferential flow and transport.
Soil Sci. Soc. Am. J. 74(6):1929–1937. doi:10.2136/sssaj2010.0 062
Mees, F. 20 03. Applications of X-ray computed tomography in the geo-
sciences. Geo l. Soc. Lo nd. 215:1–6.
Mooney, S. 2002. Three-d imensional visualizati on and quantification of
soil macroporosity and water flow patte rns using computed to mog-
raphy. Soil U se Manage. 18(2):142–151. d oi :1 0 .1111 / j .14 7 5 - 2 74 3 . 2 0 0 2 .
tb00232.x
Mooney, S. J., and C. Morris. 2 00 8. Mo rpholog ical ap proach to under-
stand ing preferent ial flow usi ng image ana lysis wit h dye tracers a nd
X-ray co mputed tomog raphy. Catena 73(2):204 –211. doi:10.1016/j.
catena.2007.09.003
Morris, C ., S.J. M ooney, and S.D. Young. 2 00 8. Sorptio n and desorpt ion
characteri stics of the dye t racer, Brilliant B lue FCF, in sandy a nd clay
soil s. Geod erm a 146:434 –438. doi:10.1016/j.geoderma.2008.06.021
Naveed, M ., P. Moldru p, H.-J. Vogel, M. Lamande, D. Wi ldenschi ld, M.
Tulle r, and L.W. de Jonge. 2014. Impact of long-ter m fer tilizati on
practi ce on soi l str uctu re evolut ion. Geoderma 217–218:181–189.
doi :10.1016/j .g eod er ma.2 013.12.0 01
Nimm o, J.R. 2010. Theor y for source -res pons ive and f ree- sur face
film model ing of u nsatu rated flow. Vadose Zone J. 9(2):295– 306.
doi:10.2136/vzj2009.0085
Nimm o, J.R. 2012. Preferenti al flow occurs in unsaturated condition s.
Hydrol . Proces ses 26(5):786 –789. doi:10.1002 /hyp.838 0
Persson, M., S. Haridy, J. Ols son, and J. Wendt. 2005. Solute trans por t
dynamics by high-reso luti on dye tracer experi ments— Imag e anal -
ysis and time moments. Vadose Zone J. 4(3):8 56–865. d o i:1 0. 213 6/
vz j20 0 4. 012 9
Rab, M.A., R.E. Hali ng, S. R. Aaron s, M. Ha nnah, I.M. Young, and D. Gib-
son. 2014. Evaluation of X- ray computed tomograp hy for quantif y-
ing macroporosit y of loamy pasture soi ls. Geoder ma 213: 460 –470.
Samma rtino, S., E. Miche l, and Y. Capowiez. 2 012. A novel method to
visua lize and cha racterize p referential fl ow in undist urbed soil cores
by using multisli ce helical CT. Vadose Zone J. 11(1). doi:10. 2136/
vz j2 011.0100
Scheibe, T.D., W.A. Pe rki ns, M.C. Richmond, M .I. McKin ley, P.D.J. Ro me-
ro-Gomez, M. Oostrom, T.W. Wiets ma, J. A. Ser kowsk i, and J.M.
Zachara . 2015. Pore- scale and multiscale numerical si mulation of
flow and trans por t in a laborato ry-scal e column. Water Resour. Res.
51(2):1023–10 35 . do i:10 .10 0 2/ 2014WR015 959
Serra, J. 1982. Image analysis and mathematical morphology. Acad.
Press, London.
Simunek, J., M.Th. van G enuchten, and M . Sejna. 20 08. D evelopment
and app lications of the HYD RUS and STANMOD soft ware pack-
ages an d related codes. Vadose Zone J. 7(2):587–60 0. d o i:1 0. 2 13 6/
vzj2007.007 7
Soil le, P. 2004. M orphological image analysis —P rincipl es and applica-
tions. 2nd ed. Spri nger-Verlag, Berlin.
Stampan oni, M ., A. Groso, A. Isenegger, G. Mikulj an, Q. Che n, and A.
Bertrand. 2006. Trends in synchrot ron- base d tomog raphic imag-
ing: The SLS experience. Proc. SPIE Int. Soc. Opt. Eng. 6318:63180M-
1– 6 318 0M -14.
Tipp kötter, R., T. Ei ckho rst, H. Taubner, B. Gredner, and G. Rade maker.
200 9. Detection of soil water in ma cropo res of undist urbed soil us-
ing microfocus X-ray t ube com puter ized to mography (mCT). Soil
Tillage Res. 105(1):12–20. doi:10.1016/j.still.2009.05.001
Vogel, H.-J., I. Cousin, O. Ip pisch, and P. Bastia n. 20 06. The dominant
role of structure for solute t ranspor t in soi l: Ex perimenta l evidence
and mod elling of structure an d transpor t in a fi eld ex periment. Hy-
drol. E arth Syst. Sci. 10(4):495– 506 . doi:10.5194/hess-10-495-2006
Vogel, H., a nd A. Kretzschma r. 1996. Topolog ical ch aracte rization of
pore space in soil—Sample p reparation a nd dig ital image -pro cess-
ing. Ge oder ma 73(1–2):23 –38. doi:10.1016/0016-7061(96)00043-2
Vogel, H.-J., U. Weller, and O. Ipp isch. 2 010. Non-equili brium in soil hy-
draulic modelling. J. Hydrol. 393:20–28.
Weiler, M., and H. Fluhl er. 20 04. Inferring flow types from dye pat tern s in
macroporous soil s. Geoderm a 120(1–2):137–153. do i:10.1016/j.ge o-
derma.2003.0 8.014
Wild ensch ild, D., and A. P. Sheppard. 2 013. X-ray imagi ng and analysis
techni ques fo r quant ify ing po re-scale structure and processes in
subsu rface porous medium sys tems. Adv. Water Resour. 51:217–246.
doi :10.1016/j .adv wat res.2012. 07.018
Youssef, S., D. Ba uer, S. Bekri, E. Rosenberg, and O. Viz ika- Kawadias.
2010. 3D In -situ flu id distributi on ima ging at the po re scale as a
new tool fo r mult ipha se flow s tudi es. SPE Annu. Tech. Conf. Exhib.
5:4271–4282.
