Geoderma 261 (2016) 178189

Contents lists available at ScienceDirect

Geoderma
journal homepage: www.elsevier.com/locate/geoderma

Mapping peat layer properties with multi-coil offset electromagnetic

induction and laser scanning elevation data
D. Altdorff a,, M. Bechtold b, J. van der Kruk a, H. Vereecken a, J.A. Huisman a
a
b

IBG 3 Agrosphere, Forschungszentrum Jlich GmbH, Jlich, Germany

Thnen Institute of Climate-Smart Agriculture, Bundesallee 50, Braunschweig, Germany

a r t i c l e

i n f o

Article history:
Received 19 March 2015
Received in revised form 20 July 2015
Accepted 21 July 2015
Available online 6 August 2015
Keywords:
Peatland
Bog
Digital soil mapping
Carbon stocks
Soil organic carbon
Bulk density
Electromagnetic induction (EMI)

a b s t r a c t
Peatlands store large amounts of soil organic carbon (SOC). Depending on their present condition, they act as a
source or sink of carbon dioxide. Therefore, peatlands are highly relevant for climate change investigations and
there is considerable interest to assess spatial heterogeneity of peat soil properties in order to assess the total
amount of stored carbon. However, reliable information about peat properties remains difcult to obtain at the
eld scale. A potential way to acquire this information is the indirect mapping of easily recordable physical variables that correlate with peat properties, such as the apparent electrical conductivity (ECa). In this study, we aim
to explore the potential of multi-coil offset electromagnetic induction (EMI) measurements to provide spatial estimates of SOC content, bulk density, and SOC stock for a highly variable and disturbed peatland relict (~35 ha)
with a remaining peat layer thickness of less than 1 m. EMI measurements comprised six integral depths that varied from 00.25 to 01.80 m. In combination with ancillary laser-scanning elevation data, a multiple linear regression model was calibrated to reference data from 34 soil cores that were used to calculate integral
properties of the upper 0.25, 0.5, and 1 m layer, as well as for the total peat layer. Leave-one-out crossvalidation for the different depth ranges resulted in a root mean square error of prediction (RMSEP) between
1.36 and 5.16% for SOC content, between 0.108 and 0.183 g cm3 for bulk density, and between 3.56 and
9.73 kg m2 for SOC stocks, which corresponds to roughly 15%, 10%, and 20% of the total eld variability, respectively. The selection of explanatory variables in the regression models showed that the EMI data were important
for accurate model predictions, while the topography-based variables mainly acted as noise suppressors. The accuracy of the SOC content estimates roughly equalled the quality of SOC content predictions obtained in previous
eld applications of the visible-near infrared technique (vis-NIR). The spatial variation of the predicted peat layer
properties showed similarities to the former land use distribution. Overall, it was concluded that EMI measurements offer a useful alternative to the established vis-NIR method for SOC content mapping in carbon-rich soils.
2015 Published by Elsevier B.V.

1. Introduction
Peatlands store large amounts of soil organic carbon (SOC). Whether
a peatland still acts as a sink for carbon dioxide (CO2) or has turned into
a CO2 source depends on its condition, i.e. its degree of disturbance due
to changing environmental conditions or land use. In particular, decreasing water levels in peatland enhance decomposition by aerobic microbial activity, which makes the carbon stock in peatland highly
vulnerable. Therefore, peatlands are of high relevance for future greenhouse gas emission and climate change predictions (Frolking et al.,
2006; Drsler et al., 2008; Kchy et al., 2015). If decomposition of the
aerated peat layer occurs, its physical and biogeochemical properties
change, which often leads to a high spatial variability of peat properties
in disturbed peatlands (Succow and Joosten, 2001; Holden, 2005).
Corresponding author at: Boreal Ecosystem Research Initiative, Grenfell Campus
Memorial University, NL, Canada.
E-mail address: d.altdorff@fz-juelich.de (D. Altdorff).

http://dx.doi.org/10.1016/j.geoderma.2015.07.015
0016-7061/ 2015 Published by Elsevier B.V.

There is considerable interest to assess spatial heterogeneity of peat

soils and the amount of carbon stored in these ecosystems across various scales. For example, current efforts to mitigate greenhouse gas
emissions from peatlands by rewetting formerly drained peatlands in
the course of restoration projects benet from spatial estimates of
peat properties and remaining SOC stocks. Unfortunately, reliable information about peat properties at the eld-scale remains difcult to obtain, since soil coring is laborious and only provides point information.
A potential way to acquire more adequate eld-scale information is provided by non-invasive mapping of easily recordable physical variables
(proxies) that correlate with relevant soil properties, such as soil
color, dielectric permittivity (), and bulk electrical conductivity (ECb).
An established sensing method for soil color is the visible-near infrared technique (vis-NIR). It is based on the light absorbance of soil components in the vis-NIR range of the electromagnetic spectrum (Stoner
and Baumgardner, 1981; Viscarra Rossel et al., 2006). In the past two decades, vis-NIR has been used for the assessment of several soil properties like nitrogen (Chang et al., 2001), potassium (Daniel et al., 2003),

D. Altdorff et al. / Geoderma 261 (2016) 178189

and clay content (Cozzolino and Moron, 2003). Various studies have
also used vis-NIR for SOC content estimation and found that vis-NIR is
a proper tool for in situ SOC content characterization (Walvoort and
McBratney, 2001; Tian et al., 2013; Debaene et al., 2014). However,
vis-NIR measurements using on-the-go sensors typically provide information at a single, and usually very shallow, measurement depth only.
Acquiring information on deeper peat characteristics can only be
achieved by soil probing, which impedes the mapping of large areas.
Geophysical methods like ground penetrating radar (GPR) and electromagnetic induction (EMI) are generally suitable to obtain information from deeper regions by non-invasively measuring and ECb,
respectively. Both variables depend on soil water content and porosity,
and the ECb is additionally inuenced by the pore water electrical conductivity (w) and the electrical conductivity along particle surfaces
(surface conductivity). These dependencies are described by available
petrophysical models (Archie, 1942; Waxman and Smits, 1968;
Rhoades et al., 1976; Topp et al., 1980). GPR is a geophysical method
that provides measurements of that can be used to estimate soil properties (Collins and Doolittle, 1987; Huisman et al., 2003; Neal, 2004).
Several studies have shown the applicability of GPR in peatland since
changes in of the different layers potentially allow the identication
of peat thickness above a mineral basement (Slater and Revee, 2002;
Kettridge et al., 2008; Proulx-McInnis et al., 2013; Comas et al., 2015).
EMI is a popular, non-invasive geophysical method to map ECb of
relatively large areas in comparably short times (e.g. Altdorff and
Dietrich, 2012; Doolittle and Brevik, 2014). It provides integral apparent
electrical conductivity (ECa) values that constitute a non-linear average
of ECb over a specic depth range (McNeill, 1980). EMI has been used to
characterize the spatial variation of soil texture (Corwin and Lesch,
2005), clay content (Weller et al., 2007), soil salinity (Rhoades, 1993),
and soil water content (Robinson et al., 2009). Although well established, EMI has rarely been used to study peat layer properties. Slater
and Revee (2002) supported their GPR study with EMI data of one
depth integral (06 m). Recent advances in technology and interpretation make EMI even more attractive for characterization of peatland.
Multi-coil offset EMI systems have become available that now allow
the simultaneous recording of several integral depths (Delefortrie
et al., 2014b; Doolittle and Brevik, 2014) and facilitate the accurate determination of ECb for several layers using inversion (Monteiro Santos
et al., 2011; von Hebel et al., 2014). In addition, readily available GPS
technology now allows the rapid acquisition of georeferenced ECa data
with a high spatial resolution. Thus, multi-coil offset EMI systems offer
the potential to resolve vertical and lateral heterogeneities (at a scale
of b1 m) of peatland at the eld scale and beyond (areas N 10 ha).
Most geophysical studies focused on intact peatlands with peat
thicknesses of up to several meters. Disturbed and drained peatlands
for which the peat properties strongly differ from the ones of natural
peatlands due to degradation processes have rarely been investigated
with geophysical methods (Walter et al., 2015). Due to the spatially
often patchy land use history including peat cutting, the peat thickness
and spatial pattern of such disturbed peatland typically shows a much
higher lateral variability than intact peatlands (De Smedt et al., 2013).
In addition, peat degradation is accompanied by changes of peat electrical properties, which may need to be considered in the interpretation of
EMI survey results. For instance, the high specic surface area of peat
soils is often associated with a high cation exchange capacity (CEC)
(Bunt, 1988), which is expected to increase with progressing peat degradation (Puustjrvi, 1956). Since CEC is positively correlated with the
amount of surface conductivity (de Lima and Niwas, 2000; Comas and
Slater, 2004), ECb is expected to increase with increasing peat degradation. Furthermore, Comas and Slater (2004) found that w was positively correlated with CEC in peat, which is expected to further increase ECb
(Archie, 1942). Simultaneously, peat degradation leads to higher soil
bulk density, lower saturated water contents, and modication of the
water retention and hydraulic conductivity characteristics (Brevik and
Fenton, 2004; Gnatowski et al., 2009; Dettmann et al., 2014; Islam

179

et al., 2014), which all affect ECb too. Walter et al. (2015) investigated
various factors that inuence ECb in a comprehensive multi-site study
and found that pore water electrical conductivity and CEC mainly controlled inter-site variability, whereas water content and CEC were
most inuential when a single site was considered. As water content
and CEC are related to peat properties like bulk density, SOC content,
and degree of decomposition, these results indicated that the mapping
of horizontal and vertical peat variability may be possible with a sitespecic calibration to ECb (Walter et al., 2015). Nevertheless, the complexity of peatlands poses several challenges for mapping peat properties with EMI. Case studies are needed to investigate the capability of
EMI to provide meaningful information about peat properties despite
potentially complex peat internal layering with abrupt and gradual
layer interfaces, sublayer anisotropy and dynamical pore water chemistry as well as the effect of various geological layers that can underlie a
peatland.
In this study, we aim to explore the potential of multi-coil offset EMI
measurements to quantify spatial variability of peat layer properties in a
disturbed peatland relict that experienced drainage, peat cutting, and
agricultural land use. Reference data from 34 cores and multiple linear
regression (MLR) analysis is used to investigate whether EMI measurements of six integral depths between 00.25 and 01.80 m can be used
to obtain information on the spatial variability of peat properties. Compared to previous geophysical studies in peatland, the SOC content and
bulk density of the peat layer at our eld site is spatially highly variable.
Thus, knowing the spatial distribution of peat thickness is not sufcient
to estimate SOC stock, which is the most relevant property for estimating potential climate change impacts. Given the spatial heterogeneity of
peat properties at our eld site, we decided to directly estimate SOC
content, bulk density, and SOC stocks instead of peat thickness. Soil
properties are known to correlate with topography-based variables
like e.g. elevation, slope and curvature, and thus data from digital elevation models are often used as ancillary data in digital soil mapping
(McBratney et al., 2003). Here, we want to evaluate the contribution
of such data to the geophysical mapping of peat properties and included
topographic information from laser scanning elevation data as ancillary
variables in our MLR analysis.

2. Methods
2.1. Site description
The test site is a 35 ha part of the disturbed bog peat complex Groes
Moor (Great Peat Bog) in northern Germany (Gifhorn, N 523454.22,
E 103946.43). The bog peat complex developed on glacial sand. During the 19th and 20th century, the bog was drained and used for peat
cutting and intensive land use (Fig. 1). Due to this history, a peat layer
of less than 1 m thickness remained from the originally up to six meters
thick peat. From the peat cutting history, elongated peat shoulders and
depressions remained. The elevation of the test site ranges from 58.01 to
60.05 m a.s.l. with a general trend of increasing elevation towards the
north-east (Fig. 2).
In the period of arable land use, parts of the test site were
plowed, which mixed peat with the underlying sand to various
degrees. Currently, the test site is used as extensive grassland with
a relatively shallow mean water table depth of about 10 to 40 cm
below the soil surface. The vegetation varies from sedge (Carex
nigra, Carex leporina) and grass (Poa pratensis, Festuca ovina,
Molinia caerulea, Eriophorum angustifolium, Juncus effusus) to moss
(Sphagnum cuspidatum, Sphagnum fallax) dominance as a function
of spatial differences in water level. Sheep grazing occurs one to
three times a year and the grass is mulched every autumn. Fertilizers are not applied (Leiber-Sauheitl et al., 2014). The spatial variation in greenhouse gas emissions at the test site was recently
investigated by Leiber-Sauheitl et al. (2014).

180

D. Altdorff et al. / Geoderma 261 (2016) 178189

Fig. 1. Schematic overview of the site and its land use history based on historical information and eld observations. A peat extraction period around 1900 was followed by intensive land
use around 1980. Today, the whole area is shallowly drained (ditches indicated by black lines) and used as extensive grassland.

2.2. EMI data collection and processing

EMI uses a transmitter coil to generate a primary magnetic eld that
induces eddy currents within the subsurface, which in turn generate a
secondary magnetic eld. The ratio of the primary to the secondary
magnetic eld measured at the receiver coils can be related to the ECb
of the subsurface (Keller and Frischknecht, 1966). EMI provides ECa
measurements that represent certain integral depth ranges that depend
on the coil distance and orientation (McNeill, 1980). In this study, we
used two identically constructed CMD-MiniExplorers (GF instruments,
Czech Republic) with different operating frequencies (22 kHz and
30 kHz) to avoid interference. Each instrument contains one transmitter
and three receiver coils with different separation from the transmitter
coil. One device was used in the vertical dipole orientation and the
other device was used in the horizontal dipole orientation. The horizontal dipole orientation has a higher sensitivity at shallow depths, while
the vertical dipole orientation has higher sensitivity in the deeper soil.
The combination of three coil separations and two dipole orientations
results in six different so-called pseudo-depths (PD). Here, PD is the approximate depth region that dominantly inuences the measured signal. In a homogeneous soil, 75% of the signal strength originates above
the PD (McNeill, 1980), which is a commonly used arbitrary threshold.
For our EMI set-up, the PDs for the horizontal dipole orientation were
25 cm, 50 cm and 90 cm, and the PDs for the vertical dipole orientation
were 50 cm, 105 cm, and 180 cm. Since vertical averaging is different for

the different dipole orientations, the weighted average of the horizontal

dipole orientation that is biased to the shallow soil is referred to with
the subscript s (i.e. PD50s) in the following, whereas the weighted
average for the vertical dipole orientation is referred to with the subscript d (i.e. PD50d).
Before EMI data acquisition, the instruments were acclimated to the
outside temperature for 20 min. The two instruments were installed on
two identical custom-built plastic sledges and pulled with an allterrain-vehicle at a speed between 7 and 10 km/h. Measurements
were acquired with a sampling rate of 10 Hz, which resulted in approximately one reading every 0.25 m. The average track separation was
~ 4 m. Each instrument was connected to a separate custom-made
single-frequency global positioning system (GPS) using a NovAtel
antenna with the standard positioning service with an accuracy of
~ 1 m. After data acquisition, measurements were interpolated on a
common 10 10 m grid using variogram analysis and ordinary block
Kriging using Surfer 8 (Golden Software, USA).
A rst preliminary EMI survey of the test site was performed in
September 2012 using one EMI instrument in the vertical orientation.
The extended EMI campaign was performed in June 2013. There was a period with low precipitation of 10 mm in the two weeks prior to the second
campaign. Water table depths in the boreholes of the soil cores ranged
from 0.11 to 0.52 m below surface during the extended campaign.
EMI devices are known to be sensitive to external inuences, such as
different operator or operating systems (sled) and changes in instrument height which makes the acquisition of reproducible and accurate
ECa data difcult (Santos and Porsani, 2011; Altdorff and Dietrich,
2014; Delefortrie et al., 2014a). Therefore, an instrument- and sitespecic calibration procedure is required (Mester et al., 2011), particularly when the instrument records physically implausible negative
values (von Hebel et al., 2014). Here, we calibrated measured ECa by
means of reference ERT proles during the full EMI campaign as suggested by several authors (Lavou et al., 2010; Mester et al., 2011;
Minsley et al., 2012; von Hebel et al., 2014). To this end, three ERT transects of 30 m were measured at different locations to capture the variability in ECb at the test site.
2.3. Soil cores and sampling design

Fig. 2. Digital elevation model (DEM) of the study site. The elongated structure of higher
elevation corresponds to the most prominent peat shoulder (indicated by white arrow).

We took 34 soil cores down to a depth of 1 m, with a diameter of

0.06 m (Geotool probing equipment, Berlin, Germany) during two drilling campaigns. The rst 15 soil cores were taken randomly within the
eastern part of the test site in September 2012, while the locations of
the remaining 19 cores were selected using the results of the preliminary EMI campaign. To this end, we divided the ECa data ranges that
were measured with the three coil separations during the rst campaign into three equally-sized data range classes representing low,

D. Altdorff et al. / Geoderma 261 (2016) 178189

intermediate, and high ECa values. This resulted in 27 possible combinations of soil electrical conductivity classes at each location. However, not
all possible combinations occurred at the test site and some classes occurred only sporadically. Therefore, we only used classes that occurred
at more than 1% of the measurement locations, which resulted in 15
sample classes. For each of these 15 classes, the location of the soil
core was determined by selecting the median of the middle coil separation. One additional soil core location was manually dened within the
area of each of the four largest classes to further increase the amount of
appropriate reference soil samples. The positions of the 19 selected soil
core sampling locations were determined in the eld using a conventional portable GPS system (Garmin etrex, USA).
2.4. Sample preparation and chemical analysis
Each peat core was divided into several samples consisting of 0.05 to
0.20 m thick layers depending on the peat stratigraphy and the depth of
the peatsand interface. The peatsand interface was more gradual in
the areas that were plowed in the past. Sand could be found in the
peat layer and some peat pieces could be found in the sand layer. However, the interface between peat and sand was still distinct and clearly
identiable in the eld from color and consistency. Isolated pieces of
peat in the sand layer were attributed to the sand layer.
The moist eld samples were weighted, homogenized, and a small
subsample was separated to determine w. For this purpose, 50 g of
dry soil was mixed with 50 ml deionized water and mixed in a shaking
device for 1 h. After centrifugation (14,000 rpm), electrical conductivity
of the supernatant water was determined. After sample drying, the w
value was calculated by assuming that electrical conductivity was linearly reduced by dilution. The remaining sample material was dried at
40 C for two weeks to determine bulk density and soil water content.
Dried samples were sieved and after mixing, subsamples of about 50 g
were ground. Approximately 0.5 g of the ground subsamples were analyzed for SOC content by dry combustion (LECO TruMac CN; LECO
Corporation, St. Joseph, Michigan, USA). Using the prole information,
values of SOC content, bulk density, soil water content, w, and SOC
stock were calculated for the depths 025, 050, and 0100 cm, as well
as for the peat layer with variable thickness.
2.5. Map of peat thickness
A digital elevation model (DEM) of the test site with a 0.5 0.5 m
resolution was determined using highly accurate laser scanning data
(N10 reections per m2) acquired in 2012. The data acquisition and processing was performed by Milan Geoservice GmbH in cooperation with
the Regional Authorities for Geo-Information and Land Development of
Lower Saxony (LGLN), Hannover, Germany. The absolute heights of the
peatsand interface determined from the soil cores were spatially interpolated by ordinary Kriging using Surfer 8 (Golden Software, USA). An
estimate of the spatial distribution of the peat thickness was obtained
by subtracting the interpolated height of the peatsand interface from
the DEM.
2.6. Multiple linear regression analysis
To predict SOC content, bulk density, and SOC stocks, we applied a
stepwise (backward) Multiple Linear Regression (MLR) approach.
MLR is an established statistical method for the generation of simple
prediction models (e.g. Backhaus et al., 2008). In backward stepwise
MLR, the available explanatory variables are successively removed at
each step when their exclusion leads to an insignicant decrease in
model t (Mundry and Nunn, 2009). We used two types of explanatory
variables in MLR: (i) EMI-based variables and (ii) topography-based
variables. For the EMI-based variables, we considered the six EMI
congurations with different PD (25, 50s, 90, 50d, 105, and 180 cm)
and a selection of ratios of these EMI congurations (so-called prole

181

ratios, PR) that are known to better reect soil heterogeneity with
depth (Corwin et al., 2003; Cockx et al., 2007; Popp et al., 2013). Here,
we considered the prole ratios PD90/PD50s, PD50s/PD25, PD180/
PD105 and PD105/PD50d. Furthermore, we also considered the standard deviation of all EMI signals (SDall) recorded at the same location
as well as the standard deviation of the three signals for the two sensor
orientations separately assuming that the horizontal dipole orientation
represents the heterogeneity in the shallow soil (SDs) and the vertical
dipole orientation represents the heterogeneity in the deeper soil (SDd).
As topography-based variables, we used elevation, slope, aspect,
prole curvature, plan curvature, and tangential curvature. These attributes were derived from the high-resolution DEM using Surfer 11
(Golden Software, USA). We also considered Northing and Easting as
explanatory variables to account for possible eld-scale trends.
We further explored the value of grouping the explanatory variables (except Northing and Easting) by means of a K-mean cluster
algorithm (MacQueen, 1967) that considered relative weights using
the Mahalanobis distance (e.g. Backhaus et al., 2008; Popp et al., 2013).
For this, we clustered the data in two groups and used this binary information as an additional explanatory variable, hereafter referred to as
cluster. The idea behind this is that the clustering indicates an overall
systematic afliation of locations within each cluster, which is potentially
also reected in the peat properties that we would like to predict.
First, all available explanatory variables were used in the backward
MLR and successively removed as described above. The backward MLR
was repeated using only the EMI-based variables in addition to
Northing (N) and Easting (E) to assess the importance of the
absence of topography-based variables for prediction accuracy. To assess
the role of selected explanatory variables in the models, we analyzed the
zero-order correlations (Pearson coefcient R) to the target variables. If
an explanatory variable is not or only very weakly correlated to the target variable, while its value within the MLR model is signicant, the explanatory variable is said to act like a so-called suppressor variable
(Nathans et al., 2012). A suppressor variable contributes to the regression analysis by removing irrelevant variance from other explanatory
variables (Nathans et al., 2012).
For each regression model, we calculated the coefcient of determination (R2) and the root mean square error (RMSE) of the calibration. The quality of all regression models was also assessed by crossvalidation using the leave-one-out approach (e.g. Picard and Cook,
1984). The performance in this validation was evaluated using the
R2P and the root mean square error of prediction (RMSEP), which
were calculated from the measured peat properties and the corresponding predictions within the cross-validation approach. The regression and
cluster analysis in this study were performed by SYSTAT 12 (Systat Software, Inc., USA).
3. Results and discussion
3.1. Soil core analysis
The interpolated peat thickness map is presented in Fig. 3. Peat
thickness ranged from 0.20 to 0.85 m with increasing depth towards
east and south-east and towards a smaller area in the north. The thicker
peat layer in the south-east reected the land use history and the historical division in arable land and grassland (Fig. 1). The peat thickness
map also shows the prominent abandoned peat shoulder parallel to
one of the drainage ditches.
The SOC contents from the soil core analysis were highly variable
in space and decreased with depth (Table 1a). This high variability is related to the land use history. At some locations, variable amounts of
sand were mixed into the peat layer by plowing, which reduced SOC
content. Furthermore, the mean SOC content decreased with increasing
depth due to the increasing contribution of the underlying sand with
low SOC content. Contrary to SOC, the bulk density increased with
depth (Table 1b). The variability in bulk density partly reected the

182

D. Altdorff et al. / Geoderma 261 (2016) 178189

map is a rst indication that the EMI signal contains information that
may explain parts of the variability in peat layer properties. The vertical
ECa distribution indicated a continuous decrease of ECb with depth for
most areas when the results for PD25 were not considered. This suggests that the ECb of the peat layer was higher than that of the sand
layer. The observed lower ECa values for PD25 as compared to PD50s
are likely associated with the unsaturated top soil due to a longer
warm and dry period before the survey campaign.
Fig. 6b shows the vertical standard deviation (SD) derived from the
measurements in the horizontal orientation (shallow), the vertical
orientation (deep) and all PD. The vertical SD varied between 0.12 and
5.8 mS m1, which suggests stronger vertical than spatial variability
of ECa. The spatial SD variability was higher in the shallow soil (SDs)
than in the deeper soil (SDd). Since the deeper SDd signal originates
dominantly from the sand layer, the lower SDd indicates less variability
in ECb within the sand.
The EMI results suggested that it may be possible to invert layer
depth and ECb from the measured ECa data (von Hebel et al., 2014).
We have explored this in detail, but we found that this approach did
not provide meaningful estimates of peat layer depth. Possible explanations for this are the small differences in measured ECa, the heterogeneity of the peat layer, as well as the small depth range of the peat layer.
The peat thickness varied between 0.20 and 0.85 m in total, but the
majority of the samples (26 out of 34) varied between 0.20 and 0.39 m
only and 18 samples varied between 0.25 and 0.35 m. This is problematic
for a peat thickness determination from EMI inversion, particularly since
the applied ECa integrals represent mostly larger soil volumes.

Fig. 3. Interpolated map of peat thickness with location of the soil cores. The absolute
heights of the peatsand interface were determined at the soil core locations and spatially
interpolated. The interpolated peatsand interface was subtracted from the digital elevation model to obtain the peat thickness map.

heterogeneity of the peat thickness, since peat is generally characterized

by higher SOC content and lower bulk density compared to the underlying sand layers. Thus, bulk density increased quickly with depth at locations with a thin peat layer. Accordingly, SOC content and bulk density
were well correlated (Fig. 4a).
Measured SOC stock variability was also high in space and with
depth (Table 1c) as a result of variation in SOC content, bulk density,
and peat thickness. Measured SOC stocks increased with peat layer
thickness, as shown in Fig. 4b. Thus, the highest SOC stocks were measured in the eastern and south-eastern part of the test site. However,
the relation in Fig. 4b is not strictly linear due to the spatial variability
in SOC content and bulk density of the peat.
Fig. 5 shows boxplots of soil water content and w of the peat and
sand layer determined for the soil cores in the second sampling campaign. The soil water content in the peat layer was almost twice as
high as in the sand layer, while w was on average 40% lower in the
peat as compared to the sand. As the differences in both variables
have an opposing effect on ECb, it cannot be concluded from the reference data that a general measurable difference in ECb exists for the
two layers over the whole site.

3.3. Results of stepwise multiple linear regression analysis

The direct correlation between the explanatory and target variables is
provided in Tables 2 and 3. Generally, the EMI-based variables (Table 2)
correlated better with the target variables than the topography-based
variables (Table 3). Furthermore, Northing and slope were moderately
correlated to the majority of target properties. In contrast, Easting,
SDall, the binary variable from the cluster analysis, and the remaining
topography-based variables were not correlated signicantly to any of
the target variables.
An overview of the stepwise MLR analyses using the full set of
explanatory variables, their variable selection, calibration and crossvalidation performance is provided in Table 5. The sets of selected
explanatory variables indicate that the vertical heterogeneity (PR
and SD) contained valuable information to predict the selected target
variables since they were selected in every model. This conrmed the
assumption that the PR and SD values were linked to the vertical
variability of the peat layer. Furthermore, all models except the
SOC content model for the 0100 cm depth range used both EMIbased and topography-based explanatory variables.
Table 4 presents an overview of the stepwise MLR analyses for the
models generated by EMI-based variables and coordinates only. As in
the analysis that included topography-based variables, all models except
the SOC stock model for the 025 cm depth range contained at least
one variable related to vertical heterogeneity in measured ECa (PR and
SD). However, the prediction accuracy decreased considerably for all

3.2. EMI survey results

The EMI survey results are shown in Fig. 6a. Overall, the observed ECa
was low and showed a relatively low spatial variability for all PD compared to other EMI studies (e.g., Abdu et al., 2008; Robinson et al.,
2009). All ECa maps show a similar eld-scale pattern with a large
area in the middle with low ECa and areas in the north and south-east
with higher ECa. This pattern shows similarities with the interpolated
peat thickness map and the land use history (Figs. 1 and 3). In addition,
an elongated structure of higher ECa appeared in all ECa maps, which is
particularly prominent in the PD50 map. This structure corresponds to
the abandoned peat shoulder that was also evident from the peat
layer thickness map (Fig. 3). The visual match with the peat thickness

Table 1
Soil properties from soil core analysis. Minimum (MIN), maximum (MAX) mean values (MEAN), and data range (RANGE) of a) soil organic carbon (SOC) content, b) bulk density, and
c) SOC stock for the upper 25, 50 and 100 cm, and the variable peat layer.

MIN
MAX
MEAN
RANGE

a)

b)

c)

SOC content [%]

Bulk density [g cm3]

SOC stock [kg m2]

25 cm

50 cm

100 cm

peat

25 cm

50 cm

100 cm

peat

25 cm

50 cm

100 cm

peat

3.0
33.2
15.9
30.2

2.0
36.7
10.0
34.7

1.1
10.2
3.0
9.1

2.9
33.4
16.0
30.5

0.3
1.3
0.6
1.0

0.2
1.5
0.9
1.3

0.6
1.9
1.4
1.3

0.2
1.3
0.6
1.1

9.8
33.6
20.8
23.8

13.9
48.5
30.9
34.6

16.3
62.0
35.9
45.7

13.9
61.4
30.6
47.5

D. Altdorff et al. / Geoderma 261 (2016) 178189

183

Fig. 4. a) Relation of bulk density and soil organic carbon (SOC) content of the different depth ranges and b) relation of peat layer thickness to the SOC stock of the upper 1 m.

target variables except the SOC content model for the 0100 cm depth
range despite the low correlation between target and topography-based
explanatory variables (Table 3). In particular, the prediction quality of
the models for SOC content and SOC stock was considerably lower. This
indicates that the topography-based variables act as noise suppressors
in the MLR analysis. A possible explanation for the noise suppressing effect of the topography-based variables is given by its inuence on the
ECa readings during the survey, i.e. micro-relief effects on the EMI instrument height and tilt during the survey. For example, an uplifting of the instrument reduces the signal response (Morris, 2009), while different
heights of the coils (e.g. due to instrument tilt) also affect the signal response (Thiesson et al., 2014). Since these effects are stronger for shallow
EMI measurements, this may explain why topography-based variables
are often selected for the prediction of swallow target variables.
Furthermore, there is a direct inuence of topography on peat soil
moisture. Local depressions and concave curvatures (PC, PLC, TC) indicate areas in which the soil surface is closer to the rather at water

table, which results in higher capillary rise, lower water tension, and
thus higher soil water content. This may lead to locally increased ECa
values. High ECa also stand for high SOC content and low bulk density
(Table 2). The topography-based variables may have improved the MLR
models as they allowed differentiating between topography- and
peat-related effects on ECa.
3.4. Accuracy of model predictions
The cross-validation results of the models presented in Table 4
are presented in Fig. 7. The RMSEP of the SOC content predictions
varied between 1.36 and 5.16% for the different depth integrals,
which is about 15% of the total SOC content range. All models provided an R2P above 0.55 except for the SOC content model for the
0100 cm depth range (Fig. 7a). The lower accuracy of this latter
model is likely related to the smaller data range for this depth
while the largest data range for the SOC content of the 050 cm

Fig. 5. Boxplots of the volumetric soil water content and pore water electrical conductivity (w) of all peat and sand samples.

184

D. Altdorff et al. / Geoderma 261 (2016) 178189

Fig. 6. Results from the electromagnetic induction (EMI) measurements, where a) the upper two rows show the apparent electrical conductivity (ECa) from the pseudo-depths (PD) of the
different coil separations and congurations, and b) the lower row shows the vertical standard deviation of the ECa of the horizontal dipole (SDs), the vertical dipole (SDd), and both dipoles
together (SDall) as proxy for vertical soil heterogeneity.

depth range yielded particularly good prediction results. This emphasizes the suitability of the EMI approach for soils with higher
SOC content and variability.

The prediction accuracy for bulk density is generally higher than for
SOC content. The RMSEP varied between 0.108 and 0.183 g cm3, which
is about 10% of the total bulk density variability. R2P values ranged

Table 2
EMI-based explanatory variables and their zero-order correlation (Pearson coefcient R) to the target variables: soil organic carbon content (SOC), bulk density, and SOC stock. Explanatory
variables are: ECa signals with different pseudo depths (PD), their prole ratio (PR), vertical standard deviation shallow (SDs), vertical standard deviation deep (SDd), and vertical standard
deviation total (SDall).
EMI-based variables

SOC content

Bulk density

SOC stock

25 cm
50 cm
100 cm
Peat
25 cm
50 cm
100 cm
Peat
25 cm
50 cm
100 cm
Peat

PD 25

PD 50s

PD 90

PD 50d

PD 105

PD 180

PR90/50s

PR50s/25

PR180/105

PR90/50d

SDs

SDd

SDall

0.3
0.0
0.1
0.3
0.4
0.1
0.1
0.4
0.5
0.2
0.1
0.1

0.6
0.3
0.3
0.6
0.7
0.5
0.4
0.6
0.4
0.4
0.3
0.3

0.7
0.4
0.4
0.6
0.7
0.5
0.4
0.7
0.4
0.4
0.3
0.4

0.7
0.6
0.6
0.7
0.8
0.7
0.6
0.8
0.4
0.5
0.5
0.5

0.7
0.4
0.3
0.6
0.7
0.5
0.4
0.6
0.4
0.4
0.3
0.3

0.5
0.2
0.2
0.5
0.5
0.3
0.2
0.5
0.3
0.3
0.2
0.2

0.6
0.4
0.3
0.6
0.7
0.5
0.4
0.7
0.5
0.4
0.3
0.4

0.6
0.4
0.3
0.6
0.7
0.5
0.4
0.6
0.3
0.4
0.3
0.3

0.1
0.2
0.3
0.1
0.1
0.3
0.3
0.1
0.0
0.1
0.2
0.1

0.5
0.1
0.1
0.5
0.5
0.2
0.1
0.4
0.3
0.2
0.1
0.1

0.6
0.5
0.4
0.6
0.6
0.6
0.5
0.7
0.4
0.4
0.3
0.5

0.0
0.3
0.4
0.0
0.1
0.3
0.4
0.2
0.1
0.2
0.3
0.2

0.1
0.1
0.0
0.0
0.1
0.0
0.0
0.0
0.1
0.1
0.0
0.1

D. Altdorff et al. / Geoderma 261 (2016) 178189

185

Table 3
Topography-based explanatory variables: digital elevation model (DEM), slope, aspect, prole curvature (PC), plan curvature (PLC), tangential curvature (TC), further Easting (E), Northing
(N), and cluster (C) and their zero-order correlation (Pearson coefcient R) to the target variables soil organic carbon content (SOC), bulk density, and SOC stock.
E

SOC content

Bulk density

SOC stock

25 cm
50 cm
100 cm
Peat
25 cm
50 cm
100 cm
Peat
25 cm
50 cm
100 cm
Peat

0.2
0.1
0.1
0.2
0.1
0.2
0.2
0.1
0.1
0.1
0.1
0.1

0.1
0.5
0.5
0.1
0.1
0.4
0.5
0.2
0.0
0.2
0.4
0.4

Topography-based variables
DEM

Slope

Aspect

PC

PLC

TC

0.0
0.1
0.1
0.1
0.2
0.1
0.2
0.2
0.1
0.0
0.1
0.1

0.4
0.5
0.4
0.4
0.3
0.5
0.5
0.3
0.2
0.4
0.4
0.5

0.0
0.0
0.0
0.1
0.1
0.1
0.1
0.1
0.0
0.1
0.1
0.0

0.0
0.1
0.2
0.0
0.1
0.0
0.0
0.1
0.2
0.1
0.2
0.2

0.0
0.0
0.1
0.0
0.1
0.0
0.0
0.1
0.1
0.1
0.2
0.1

0.0
0.1
0.2
0.1
0.1
0.1
0.1
0.1
0.1
0.0
0.1
0.1

0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.1
0.0
0.0
0.1
0.0

between 0.751 and 0.824 (Fig. 7b). Coefcients in Table 2 already indicated that ECa is more strongly correlated to bulk density than to SOC.
We provide two reasons for the higher correlation of ECa to bulk density
compared to SOC. First, ECa of (nearly) saturated porous media depends
on porosity (Archie, 1942), which is better correlated to bulk density
than to SOC. There are processes that lead to different porosities (and
ECa) for the same SOC contents, as for instance compaction. This may
have weakened the relation of ECa to SOC. Second, the bulk density
data are expected to be more accurate than the SOC data. SOC content
is determined for small fractions of the sub-layers and the unavoidable
incomplete mixing of sample material before sub-sampling inevitably
reduces the accuracy of the SOC content determination. In contrast,
bulk density is determined for the whole layer by drying, which is a
very simple method that provides accurate and robust results. It is
also interesting to note here that the correlation between bulk density
and ECa is negative. This is contrary to the effect of more continuous
pathways for current ow generated by more particle-to-particle contacts at higher bulk densities (Rhoades and Corwin, 1990). We believe
that this is due to the moist conditions at our site with capillary rise
from shallow water levels. Particle-to-particle contacts are more relevant for drier soils in which the conduction through water lms and
along particle surfaces is dominant. At our site, water contents in peat
and sand are around 0.45 and 0.3, respectively, and we assume that conduction through water-lled capillaries is dominant.
The RMSEP of the SOC stock predictions ranged from 3.56 to
9.73 kg m2 (Fig. 7c), which is about 20% of the total SOC stock variability. R2P values varied between 0.249 and 0.479. The most accurate predictions were obtained for the SOC stock model for the peat layer,
followed by the SOC stock model for the 025 cm depth range. This
may indicate that the topography-based variables particularly enhanced the quality of the shallow soil predictions. The accuracy of the

stock predictions did not reach the quality of the SOC content and
bulk density models.
The performance of the MLR models can also be evaluated by the
credibility of the predicted spatial patterns. The estimated spatial pattern of the predicted SOC content, bulk density, and SOC stock for different depth ranges is presented in Fig. 8. Fig. 8a shows the spatial variation
of SOC content. The highest predicted values were located in the southeastern part and corresponded with the presence of the only weakly degraded deep peat in that area (compare Fig. 3). In addition, all maps
showed a zonation into a northern, middle and southern area, which
is especially prominent in the SOC content maps for the 025 cm
depth range and the entire peat layer. This zonation approximately corresponds to the historical land use map (Fig. 1). Lower SOC contents occurred within the area that was used as arable land because of the
mixing of peat and sand by plowing. The predicted SOC content maps
also reproduced the abandoned peat shoulder, which is particularly
clear in the SOC content map for the top 25 cm.
The predicted bulk density patterns in Fig. 8b are similar to the observed pattern for SOC content. The zone in the middle of the area is
characterized by a high bulk density, which is again associated with
plowing and subsequent mixing of sand and peat. The bulk density
also increased with depth as a consequence of the increasing sand
layer contribution as well as by the weight of the overlying material.
The bulk density predictions also indicated areas of lower bulk density
in the south and south-east, which are related to the deeper peat. Although to a lesser extent, the predictions of SOC stock in Fig. 8c also reect the land use history. Lower stocks were predicted in areas of
former arable land use. The map of the SOC stock in the entire peat
layer also partly reproduced the abandoned peat shoulders with values
up to 70 kg m 2. Furthermore, the higher SOC stocks in the southeastern part of the test site were also correctly predicted.

Table 4
Overview of the nal multiple linear regression (MLR) models after backward stepwise MLR. Quality of the models is summarized by the root mean square error (RMSE), coefcient of
determination (R2), root mean square error of the prediction (RMSEP), and R2 of the cross validation (R2P). The used variable names are dened in Tables 2 and 3.
Calibration

SOC content

Bulk density

SOC stock

25 cm
50 cm
100 cm
Peat
25 cm
50 cm
100 cm
Peat
25 cm
50 cm
100 cm
Peat

Validation

Selected variables

R2

RMSE

R2P

RMSEP

E, PD50d, PR105/50d, SDd, slope, PLC, TC

E, PD50s, 105, PR105/50d, slope, PLC, C, SDs
PD25, 50s, 90, 50d, 105, 180, PR50s/25, PR105/50d, SDs, SDd,
E, PD25, PR50s/25, PR180/105, SDall slope, TC
PD 105, PR105/50d, PR 180/105, slope, PC, PLC
PD50s, 90, 105, PR50s/25, PR105/50d, SDd, slope, PC
E, N, PD25, 50s, 50d,105, PR50s/25, SDd, DEM, PC, TC,
E, N, PD25, 50s, PR50s/25, DEM, PC, SDd
E, PD25, 50s, 180, PC, PLC, TC, SDall,
PD25, PR50s/25, DEM, slope, PLC,C, SDd, SDall,
PD50s, 90, 50d, 105, 180, PC, SDs, SDd
N, PD25, 50s, PR50s/25DEM, PC, PLC, TC

0.766
0.857
0.760
0.730
0.860
0.868
0.938
0.861
0.706
0.645
0.598
0.682

4.45
3.89
1.10
4.29
0.097
0.158
0.090
0.107
3.04
6.49
8.30
7.63

0.588
0.719
0.455
0.550
0.780
0.755
0.824
0.751
0.451
0.279
0.249
0.479

5.16
4.68
1.36
4.85
0.108
0.183
0.122
0.123
3.56
7.95
9.73
8.37

186

D. Altdorff et al. / Geoderma 261 (2016) 178189

Table 5
Overview of nal multiple linear regression (MLR) models when only using EMI-based variables for model building. Quality of the models is summarized by root mean square error
(RMSE), coefcient of determination (R2), root mean square error of the prediction (RMSEP), and R2 of the cross validation (R2P). The used variable names are dened in Tables 2 and 3.
Calibration

SOC content

Bulk density

SOC stock

25 cm
50 cm
100 cm
Peat
25 cm
50 cm
100 cm
Peat
25 cm
50 cm
100 cm
Peat

Validation

Selected variables

R2

RMSE

R2P

RMSEP

PD105, SDd
PD50s, 50d, 105, PR 105/50d, SDs
PD25, 50s, 90, 50d,105, 180 PR50s/25, 105/50s SDs, SDd
PR 105/50s, SDd
E, PD25, 50s, PR50s/25, 105/50d, SDd
N, PD90, 105, PR 50s/25, 105/50s, Sdall
E, N, PD25, 50s, 50d, 105, PR 50s/25, PR105/50d, Sdall
E, PD25, 50s, PR 50s/25, SDd
PD25
PD180, SDd
PD105, SDd
PD90, 50d, 105, 180, DSs, SDd

0.519
0.736
0.760
0.483
0.720
0.795
0.920
0.764
0.215
0.276
0.315
0.529

5.84
5.00
1.10
5.44
0.134
0.188
0.098
0.132
4.38
8.34
9.74
8.93

0.405
0.613
0.455
0.344
0.526
0.666
0.828
0.625
0.13
0.143
0.183
0.225

6.20
5.48
1.36
5.85
0.159
0.215
0.118
0.151
4.48
8.66
10.15
10.21

The reliable reproduction of the main peat characteristics inspires

condence that the applied EMI approach is suitable for mapping peat
property patterns. The quality of the SOC content predictions is similar
to the accuracy achieved with vis-NIR measurement in a similar
peatland setting (Knadel et al., 2011). When comparing the EMI and
vis-NIR approach, it should however be noted that maps from vis-NIR
typically only provide information about soil properties at a specic
depth and that the vis-NIR measurement does not integrate over a

depth integral (e.g. Huang et al., 2007; Muoz and Kravchenko, 2011).
In contrast, estimates of SOC content obtained with EMI in this study
refer to a specic depth range.
We like to emphasize that the heterogeneity of the investigated
peatland is strong and the result of several processes, starting from a heterogeneous peatland formation and followed by a spatially variable drainage, peat cutting, and land use history which included partial plowing of
the area. Spatial water table depth variation further inuenced the

Fig. 7. Multiple linear regression (MLR) prediction results by leave-one-out cross validation for the selected models given in Table 4 for a) soil organic carbon (SOC) content [%], b) bulk
density [g m3], and c) SOC stock [kg m2].

D. Altdorff et al. / Geoderma 261 (2016) 178189

187

Fig. 8. Spatial variation of the predicted a) soil organic carbon (SOC) content [%], b) bulk density [g m3], and c) SOC stock [kg m2] for different integral depth ranges.

spatially variable decomposition rates. Therefore, the presented results

can be considered as a worst-case scenario for the derivation of clear dependencies between EMI signals and peat properties.
Finally, it is important to realize that site-specic calibration will always be necessary to derive adequate models to predict SOC content,
bulk density, and SOC stocks from EMI measurements. This is related
to the various spatially and temporally variable factors that inuence
ECb. To minimize the effect of confounding factors, it is important that
EMI measurements are conducted within a short time period with
minor changes of water table depth, soil water content, and pore
water conductivity.

4. Conclusions
Our study demonstrated that multi-offset electromagnetic induction
(EMI) measurements contain valuable information to predict spatial
heterogeneity of peat layer properties. The visualization of the ECa
data measured with EMI already revealed a zonation of the area that
is related to the former land use. Together with laser scanning elevation
and soil core reference data, Multiple Linear Regression (MLR) models
were built that provided reliable predictions for SOC content, bulk density, and SOC stocks. The predicted spatial variation of these target properties also corresponded with the land use history of the peatland.
The selection of explanatory variables in the MLR analysis showed
that EMI-based explanatory variables were most important for accurate
predictions. In particular, the variables that captured vertical heterogeneity of ECa were consistently selected in all regression models. The
omission of topography-based variables decreased the prediction

accuracy considerably. Because of the low correlation between the

topography-based and target variables, this indicates that these variables acted as important noise suppressors. The important contribution
of noise suppressors in our study suggests that it may be worthwhile to
include additional spatial data to further suppress noise. This could for
instance be a vegetation map as a proxy for water table depth and salinity, which both also inuence ECb and thus potentially compromise a
clear relationship between measured ECa and SOC content, bulk density,
and SOC stocks.
The achieved prediction accuracy for SOC content of the demonstrated
study was similar to the prediction accuracy of vis-NIR measurements.
Therefore, we conclude that multi-coil offset EMI measurements are a
useful alternative to the established vis-NIR approach for SOC mapping
of organic soils, particularly when it is considered that EMI measurements
provide integrated information for several depth ranges. Moreover, the
demonstrated approach offers the possibility to directly estimate SOC
stocks at the eld scale and thus allows the assessment of the size of the
entire vulnerable carbon pool of disturbed peatland areas.

Acknowledgments
We would like to thank Ulrike Grlich, Claudia Wiese,
Brbel Tiemeyer, Daniel Ziehe, and the Thnen eld team of the
Bodenzustandserhebung (BZE) and students for eld and laboratory support. We thank Schferei Paulus and Norbert Horny
(Niederschsischer Landesbetrieb fr Wasserwirtschaft, Ksten-und
Naturschutz, Braunschweig) for site access and research permission at
the eld site. The authors also acknowledge the Terrestrial Environmental

188

D. Altdorff et al. / Geoderma 261 (2016) 178189

Observatories (TERENO) Initiative of the Helmholtz-Association for providing the technical equipment.
References
Abdu, H., Seyfried, D.A., Jones, S.B.M., 2008. Geophysical imaging of watershed subsurface
patterns and prediction of soil texture and water holding capacity. Water Resour. Res.
44 (4), W00D18.
Altdorff, D., Dietrich, P., 2012. Combination of electromagnetic induction EMI and
gamma-spectrometry using K-means clustering: a study for evaluation of site
partitioning. J. Plant Nutr. Soil Sci. 175 (3), 345354.
Altdorff, D., Dietrich, P., 2014. Delineation of areas with different temporal behavior of soil
properties at a landslide affected Alpine hillside using time-lapse electromagnetic
data. Environ. Earth Sci. 72 (5), 13571366.
Archie, G.E., 1942. The electrical resistivity log as an aid in determining some reservoir
characteristics. Pet. Trans. AIME 146, 5462.
Backhaus, K., Erichson, B., Plinke, W., Weiber, R., 2008. Multivariate Analysemethoden
Eine anwendungsorientierte Einfhrung. Springer, Berlin.
Brevik, E.C., Fenton, T.E., 2004. The effect of changes in bulk density on soil electrical conductivity as measured with the Geonics EM-38. Soil Surv. Horiz. 45, 96102.
Bunt, A.C., 1988. Media and Mixes for Container Grown Plants. Unwin Hyman, London
978-94-011-7906-5.
Chang, C.W., Laird, D.A., Mausbach, M.J., Hurburgh Jr., C.R., 2001. Near-infrared reectance
spectroscopy principal components regression analysis of soil properties. Soil Sci.
Soc. Am. J. 65 (2), 480490.
Cockx, L., Van Meirvenne, M., De Vos, B., 2007. Using the EM38DD soil sensor to delineate
clay lenses in a sandy forest soil. Soil Sci. Soc. Am. J. 71 (4), 13141322.
Collins, M.E., Doolittle, J.A., 1987. Using ground-penetrating radar to study soil
microvariability. Soil Sci. Soc. Am. J. 51 (2), 491493.
Comas, X., Slater, L., 2004. Low-frequency electrical properties of peat. Water Resour. Res.
40 (12).
Comas, X., Terry, N., Slater, L., Warren, M., Kolka, R., Kristijono, A., Sudiana, N., Nurjaman,
D., Darusman, T., 2015. Imaging tropical peatlands in Indonesia using ground penetrating radar (GPR) and electrical resistivity imaging (ERI): implications for carbon
stock estimates and peat soil characterization. Biogeosciences 12, 29953007.
Corwin, D.L., Lesch, S.M., 2005. Apparent soil electrical conductivity measurements in agriculture. Comput. Electron. Agric. 46, 1143.
Corwin, D.L., Kaffka, S.R., Hopmans, J.W., Mori, Y., van Groenigen, J.W., van Kessel, C.,
Lesch, S.M., Oster, J.D., 2003. Assessment and eld-scale mapping of soil quality properties of a saline sodic soil. Geoderma 114 (34), 231259.
Cozzolino, D., Moron, A., 2003. The potential of near-infrared reectance spectroscopy to
analyse soil chemical and physical characteristics. J. Agric. Sci. 140, 6571.
Daniel, K.W., Tripathi, N.K., Honda, K., 2003. Articial neural network analysis of laboratory and in situ spectra for the estimation of macronutrients in soils of Lop Buri
Thailand. Aust. J. Soil Res. 41 (1), 4759.
de Lima, O.A.L., Niwas, S., 2000. Estimation of hydraulic parameters of shaly sandstone
aquifers from geoelectrical measurements. J. Hydrol. 235 (1), 1226.
De Smedt, P., Van Meirvenne, M., Herremans, D., De Reu, J., Saey, T., Meerschman, E.,
Crombe, P., De Clercq, W., 2013. The 3-D reconstruction of medieval wetland reclamation through electromagnetic induction survey. Sci. Rep. 3, 1517.
Debaene, G., Niedwiecki, J., Pecio, A., urek, A., 2014. Effect of the number of calibration
samples on the prediction of several soil properties at the farm-scale. Geoderma 214,
114125.
Delefortrie, S., De Smedt, P., Saey, T., Van De Vijver, E., Van Meirvenne, M., 2014a. An efcient calibration procedure for correction of drift in EMI survey data. J. Appl.
Geophys. 110, 115125.
Delefortrie, S., Saey, T., Van De Vijver, E., De Smedt, P., Missiaen, T., Demerre, T., Van
Meirvenne, M., 2014b. Frequency domain electromagnetic induction survey in the intertidal zone: limitations of low-induction-number and depth of exploration. J. Appl.
Geophys. 100, 1422.
Dettmann, U., Bechtold, M., Frahm, E., Tiemeyer, B., 2014. On the applicability of unimodal
and bimodal van Genuchten-Mualem based models to peat and other organic soils
under evaporation conditions. J. Hydrol. 515, 103115.
Doolittle, J.A., Brevik, E.C., 2014. The use of electromagnetic induction techniques in soils
studies. Geoderma 223225, 3345.
Drsler, M., Freibauer, A., Christensen, T.R., Friborg, T., 2008. Observations and status of
peatland greenhouse gas emissions in Europe. The continental-scale greenhouse
gas balance of Europe. Ecological Studies 203, pp. 243261.
Frolking, S., Roulet, N., Fuglestvedt, J., 2006. How northern peatlands inuence the Earth's
radiative budget: sustained methane emission versus sustained carbon sequestration.
J. Geophys. Res. Biogeosci. 111, G1.
Gnatowski, T., Szatylowicz, J., Brandyk, T., Kechavarzi, C., 2009. Hydraulic properties of fen
peat soils in Poland. Geoderma 154 (34), 188195.
Holden, J., 2005. Peatland hydrology and carbon release: why small-scale process matters.
Philos. Trans. R. Soc. A Math. Phys. Eng. Sci. 363, 28912913.
Huang, X.W., Senthilkurnar, S., Kravchenko, A., Thelen, K., Qi, J.G., 2007. Total carbon
mapping in glacial till soils using near-infrared spectroscopy, Landsat imagery and
topographical information. Geoderma 141 (12), 3442.
Huisman, J.A., Hubbard, S.S., Redman, J.D., Annan, A.P., 2003. Measuring soil water content
with ground penetrating radar: a review. Vadose Zone J. 2, 476491.
Islam, M.M., Meerschman, E., Saey, T., De Smedt, P., Van De Vijver, E., Delefortrie, S.,
Van Meirvenne, M., 2014. Characterizing compaction variability with an electromagnetic induction sensor in a puddled paddy rice eld. Soil Sci. Soc. Am. J. 78
(2), 579588.

Keller, G.V., Frischknecht, F.C., 1966. Electrical Methods Of Geophysical Prospecting. International Series of Monographs in Electromagnetic Waves. Pergamon Press, Oxford,
New York.
Kettridge, N., Comas, X., Baird, A., Slater, L., Strack, M., Thompson, D., Jol, H., Binley, A., 2008.
Ecohydrologically important subsurface structures in peatlands revealed by groundpenetrating radar and complex conductivity surveys. J. Geophys. Res. 113, G04030.
Knadel, M., Thomsen, A., Greve, M.H., 2011. Multisensor on-the-go mapping of soil organic carbon content. Soil Sci. Soc. Am. J. 75 (5), 17991806.
Kchy, M., Hiederer, R., Freibauer, A., 2015. Global distribution of soil organic carbon
Part 1: masses and frequency distributions of SOC stocks for the tropics, permafrost
regions, wetlands, and the world. Soil 1, 351365.
Lavou, F., van der Kruk, J., Rings, J., Andre, F., Moghadas, D., Huisman, J.A., Lambot, S.,
Weihermller, L., van der Borght, J., Vereecken, H., 2010. Electromagnetic induction
calibration using apparent electrical conductivity modelling based on electrical resistivity tomography. Near Surf. Geophys. 8 (6), 553561.
Leiber-Sauheitl, K., Fu, R., Voigt, C., Freibauer, A., 2014. High CO2 uxes from grassland on
histic Gleysol along soil carbon and drainage gradients. Biogeosciences 11, 749761.
MacQueen, J., 1967. Some methods for classication and analysis of multivariate observations. In: Le Cam, L.M., Neyman, J. (Eds.), Proceedings of the 5th Berkeley
Symposium on Mathematical Statistics and Probability 1. University of California
Press, pp. 281297.
McBratney, A.B., Santos, M.L.M., Minasny, B., 2003. On digital soil mapping. Geoderma 117
(12), 352.
McNeill, J.D., 1980. Electromagnetic terrain conductivity measurement at low induction
numbers. Technical Note TN-6. Geonics. Ltd., Mississauga, Canada.
Mester, A., van der Kruk, J., Zimmermann, E., Vereecken, H., 2011. Quantitative two-layer
conductivity inversion of multi-conguration electromagnetic induction measurements. Vadose Zone J. 10 (4), 13191330.
Minsley, B.J., Smith, B.D., Hammack, R., Sams, J.I., Veloski, G., 2012. Calibration and ltering
strategies for frequency domain electromagnetic data. J. Appl. Geophys. 80, 5666.
Monteiro Santos, F.A., Triantalis, J., Bruzgulis, K., 2011. A spatially constrained 1D inversion algorithm for quasi-3D conductivity imaging: application to DUALEM-421 data
collected in a riverine plain. Geophysics 76 (2), B43B53.
Morris, E.R., 2009. Height-above-ground effects on penetration depth and response of
electromagnetic induction soil conductivity meters. Comput. Electron. Agric. 68 (2),
150156.
Mundry, R., Nunn, C., 2009. Stepwise model tting and statistical inference: turning noise
into signal pollution. Am. Nat. 173 (1), 119123.
Muoz, J.D., Kravchenko, A., 2011. Soil carbon mapping using on-the-go near infrared
spectroscopy, topography and aerial photographs. Geoderma 166 (1), 102110.
Nathans, L.L., Oswald, F.L., Nimon, K., 2012. Interpreting multiple linear regression:
a guidebook of variable importance. Pract. Assess. Res. Eval. 17 (9), 119.
Neal, A., 2004. Ground-penetrating radar and it use in sedimentology: principles, problems and progress. Earth Sci. Rev. 66 (34), 261330.
Picard, R.R., Cook, R.D., 1984. Cross-validation of regression models. J. Am. Stat. Assoc. 79
(387), 575583.
Popp, S., Altdorff, D., Dietrich, P., 2013. Assessment of shallow subsurface characterization
with non-invasive geophysical methods at the intermediate hill-slope scale. Hydrol.
Earth Syst. Sci. 9, 25112539.
Proulx-McInnis, S., St-Hilaire, A., Rousseau, A.N., Jutras, S., 2013. A review of groundpenetrating radar studies related to peatland stratigraphy with a case study on the
determination of peat thickness in a northern boreal fen in Quebec, Canada. Prog.
Phys. Geogr. 37 (6), 767786.
Puustjrvi, V., 1956. On the cation exchange capacity of peats and on the factors of inuence upon its formation. Acta Agric. Scand. 6 (4), 410449.
Rhoades, J.D., 1993. Electrical conductivity methods for measuring and mapping soil salinity. Adv. Agron. 49, 201251.
Rhoades, J.D., Corwin, D.L., 1990. Soil electrical conductivity: effects of soil properties and
application to soil salinity appraisal. Commun. Soil Sci. Plant Anal. 21, 837860.
Rhoades, J.D., Raats, P.A.C., Prather, R.J., 1976. Effects of liquid-phase electricalconductivity, water-content, and surface conductivity on bulk soil electricalconductivity. Soil Sci. Soc. Am. J. 40 (5), 651655.
Robinson, D.A., Lebron, I., Kocar, B., Phan, K., Sampson, M., Crook, N., Fendorf, S., 2009.
Time-lapse geophysical imaging of soil moisture dynamics in tropical deltaic soils:
an aid to interpreting hydrological and geochemical processes. Water Resour. Res.
45 (4), W00D32.
Santos, R.N.S., Porsani, J.L., 2011. Comparing performance of instrumental drift correction
by linear and quadratic adjusting in inductive electromagnetic data. J. Appl. Geophys.
73 (1), 17.
Slater, L.D., Revee, A., 2002. Investigating peatland stratigraphy and hydrogeology using
integrated electrical geophysics. Geophysics 67 (2), 365378.
Stoner, E.R., Baumgardner, F., 1981. Characteristic variations in reectance of surface soils.
Soil Sci. Soc. Am. J. 45 (6), 11611165.
Succow, M., Joosten, H. (Eds.), 2001. Landschaftskologische Moorkunde.
Schweizerbart'sche, Stuttgart.
Thiesson, J., Kessouri, P., Schamper, C., Tabbagh, A., 2014. Calibration of frequency-domain
electromagnetic devices used in near-surface surveying. Near Surf. Geophys. 12 (4),
481491.
Tian, Y., Zhang, J., Yao, X., Cao, W., Zhu, Y., 2013. Laboratory assessment of three quantitative methods for estimating the organic matter content of soils in China based on
visible/near-infrared reectance spectra. Geoderma 202203, 161170.
Topp, G.C., Davis, J.L., Annan, A.P., 1980. Electromagnetic determination of soil water content:
measurements in coaxial transmission lines. Water Resour. Res. 16 (3), 574582.
Viscarra Rossel, R.A., Walvoort, D.J.J., McBratney, A.B., Janikc, L.J., Skjemstad, J.O., 2006.
Visible, near infrared, mid infrared or combined diffuse reectance spectroscopy for
simultaneous assessment of various soil properties. Geoderma 131 (12), 5975.

D. Altdorff et al. / Geoderma 261 (2016) 178189

von Hebel, C., Rudolf, S., Mester, A., Huisman, J.A., Kumbhar, P., Vereecken, H., Van der
Kruk, J., 2014. Three-dimensional imaging of subsurface structural patterns using
quantitative large-scale multi-conguration electromagnetic induction data. Water
Resour. Res. 50 (3), 27322748.
Walter, J., Lck, E., Bauriegel, A., Richter, C., Zeitz, J., 2015. Multi-scale analysis of electrical
conductivity of peatlands for the assessment of peat properties. Eur. J. Soil Sci. 66 (4),
639650.
Walvoort, D.J.J., McBratney, A.B., 2001. Diffuse reectance spectrometry as a proximal
sensing tool for precision agriculture. In: Grenier, G., Blackmore, S. (Eds.), ECPA

189

2001, Third European Conference on Precision Agriculture. Agro Montpellier,

pp. 503507.
Waxman, M.H., Smits, L.J.M., 1968. Electrical conductivities in oil-bearing shaly sands. Soc.
Pet. Eng. J. 8 (2), 107122.
Weller, U., Zipprich, M., Sommer, M., Zu Castell, W., Wehrhand, M., 2007. Mapping clay
content across boundaries at the landscape scale with electromagnetic induction.
Soil Sci. Soc. Am. J. 71 (6), 17401747.