Introduction
Soils are the largest terrestrial reservoirs of soil organic carbon (SOC),
storing 2 to 3 times as much C as the atmosphere and biosphere (Chen
et al., 2015; Lal et al., 2004). In the context of climate change mitigation
as well as soil fertility and food security, there has been considerable
interest in the development of SOC, especially in erosion-affected
agricultural landscapes (Berhe and Kleber, 2013; Conant et al., 2011;
Doetterl et al., 2016; Stockmann et al., 2015; Van Oost et al., 2007; Xiong
et al., 2016). Detecting the development of soil organic carbon stocks
(ΔSOC) in agricultural landscapes needs to consider three major
challenges: first, the high small-scale spatial heterogeneity of SOC (e.g.,
Conant et al., 2011; Xiong et al., 2016). Erosion and land use change
reinforce natural spatial and temporal variability, especially in hilly
landscapes such as hummocky ground moraines where correlation lengths in
soil parameters of 10–30 m are very common. Second, pronounced short-term
temporal dynamics, caused by, e.g., type of cover crop, frequent crop
rotation and soil cultivation practices need to be considered. Third, the rather small magnitude
of ΔSOC compared to total SOC stocks need to be considered (e.g., Conant et al., 2011;
Poeplau et al., 2016).
However, information on the development of SOC is an essential precondition
to improve the predictive ability of terrestrial C models (Luo et al.,
2016). As a result, sensitive measurement techniques are required to
precisely assess short-term temporal and small-scale (10–30 m) spatial
dynamics in ΔSOC (Batjes and van Wesemael, 2015). To date, the
assessment of ΔSOC has typically been based on two methods, namely (i)
destructive, repeated soil inventories through soil resampling and (ii)
non-destructive determination of net ecosystem C balance (NCEB) by measurements
of gaseous C exchange, C import and C export (Leifeld et al., 2011; Smith et
al., 2010).
The first method is usually used during long-term field trials (Batjes and
van Wesemael, 2015; Chen et al., 2015; Schrumpf et al., 2011). Given a
sufficient time horizon of 5 to 10 years, the soil resampling method is
generally able to reveal spatial patterns and trends within ΔSOC
(Batjes and van Wesemael, 2015; Schrumpf et al., 2011). Most repeated soil
inventories are designed to study treatment differences in the long term. As
a result, short-term temporal dynamics in C exchange remain concealed
(Poeplau et al., 2016; Schrumpf et al., 2011). A number of studies tried to
overcome this methodical limitation by increasing (e.g., to monthly) the soil
sampling frequency (Culman et al., 2013; Wuest, 2014). This allows for the
detection of seasonal patterns of ΔSOC but still mixes temporal and
spatial variability of SOC because every new soil sample represents not only
a repetition in time but also in space. Temporal differences observed
through repeated soil sampling are therefore always spatially biased.
By contrast, the NECB (Smith et al., 2010) – used as a proxy for temporal
dynamics of ΔSOC – can be easily derived through the eddy covariance
(EC) technique, representing a common approach to obtaining gaseous C exchange
(Alberti et al., 2010; Leifeld et al., 2011; Skinner and Dell, 2015).
However, C fluxes based on EC measurements are integrated over a larger,
changing footprint area (several hectares). As a result, small-scale
(< 20 m) spatial differences in NECB and ΔSOC are not
detected.
Accounting for the abovementioned methodical limitations, a number of
studies investigated spatial patterns in gaseous C exchange by using manual
chamber measurement systems (Eickenscheidt et al., 2014; Pohl et al., 2015).
Compared to EC measurements, these systems are characterized by a low
temporal resolution, where the calculated net ecosystem CO2 exchange
(NEE) is commonly based on extensive gap filling (Gomez-Casanovas et al.,
2013; Savage and Davidson, 2003) conducted using empirical modeling, for
example
(Hoffmann et al., 2015). Therefore, management practices and different
stages in plant development that are needed to precisely detect NEE often
remain unconsidered (Hoffmann et al., 2015).
Compared to previously mentioned approaches for detecting ΔSOC by either
repeated soil sampling or observations of the gaseous C exchange,
automatic chamber (AC) systems combine several advantages. On the one hand,
flux measurements of the same spatial entity avoid the mixing of spatial and
temporal variability, as done in the case of point measurements from repeated soil
inventories. On the other hand, AC measurements combine the advantages of EC and
manual chamber systems because they not only increase the temporal
resolution compared to manual chambers but also allow for the detection of
small-scale spatial differences and treatment comparisons regarding the
gaseous C exchange (Koskinen et al., 2014).
To date, hardly any direct comparisons between AC-derived C budgets and soil
resampling-based ΔSOC values have been reported in the literature.
Leifeld et al. (2011) and Verma et al. (2005) compared the results of
repeated soil inventories with EC-based C budgets over 5- and 3-year study
periods, respectively. Even though temporal dynamics in ΔSOC were
shown for grazed pastures and intensively used grasslands, for example (Skinner and
Dell, 2015; Leifeld et al., 2011), no attempt was made to additionally detect
small-scale differences in ΔSOC. In our study, we introduce the
combination of AC measurements and empirically modeled aboveground biomass
production (NPPshoot) as a precise method to detect small-scale spatial
differences and short-term temporal dynamics of NECB and thus ΔSOC.
Measurements were performed from 2010 to 2014 under a silage maize – winter fodder rye – sorghum-Sudan grass hybrid – alfalfa crop rotation at an
experimental plot located in the hummocky ground moraine landscape of
northeastern
Germany.
We hypothesize that the AC-based C budget method is able to detect
small-scale spatial and short-term temporal dynamics of NECB and thus
ΔSOC in an accurate and precise manner. Therefore, we compare
ΔSOC values measured by soil resampling with NECB values derived
through AC-based C budgets (Fig. 1).
Schematic representation of the study concept used to detect
changes in soil organic carbon stock (ΔSOC). Black stars represent
SOC measured by the soil resampling method. Black circles represent annual
NECB derived using the C budget method.
![]()
Materials and methods
Study site and experimental setup
Measurements were performed at the 6 ha experimental field “CarboZALF-D”.
The site is located in a hummocky arable soil landscape within the Uckermark
region (northeastern Germany, 53∘23′ N, 13∘47′ E, ∼ 50–60 m a.s.l.). The temperate climate is characterized by a mean annual air
temperature of 8.6 ∘C and annual precipitation of 485 mm
(1992–2012, ZALF research station, Dedelow). Typical landscape elements
vary from flat summit and depression locations with a gradient of
approximately 2 %, across longer slopes with a medium gradient of
approximately
6 %, to short and rather steep slopes with a gradient of up to 13 %.
The study site shows complex soil patterns mainly influenced by erosion and
relief and parent material, e.g., sandy to marly glacial and glaciofluvial
deposits. The soil-type inventory of the experimental site consists of
non-eroded Albic Luvisols (Cutanic) at the flat summits, strongly eroded
Calcic Luvisols (Cutanic) on the moderate slopes, extremely eroded Calcaric Regosols (Densic) on the steep slopes and a colluvial soil, i.e., Endogleyic
Colluvic Regosols (Eutric), over peat in the depression (IUSS Working Group
WRB, 2015).
Transect of automatic chambers and chamber positions within the
depression overlying the Endogleyic Colluvic Regosol (WRB 2015, left). The
black arrow shows the position of the data logger and controlling devices,
which were placed within a wooden, weather-sheltered house. The soil profile
is shown on the left. Soil horizon-specific SOC (%) and Nt (%)
contents are indicated by solid and dashed vertical white lines,
respectively. Spatial differences in NECB and the basic principle of the C
budget method are shown as the scheme within the picture.
![]()
During June 2010, four automatic chambers and a WXT520 climate station
(Vaisala, Vantaa, Finland) were set up at the depression (Sommer et al.,
2016) (see Sect. 2.2.1). The chambers were arranged along a topographic gradient
(upper (A), upper middle (B), lower middle (C) and lower (D) chamber
position; length ∼ 30 m; difference in altitude
∼ 1 m) within a distance of approximately 5 m of each other (Fig. 2). As part of the CarboZALF project, a manipulation experiment was carried
out at the end of October 2010, i.e., after the vegetation period (Deumlich
et al., 2017). Topsoil material from a neighboring hillslope was
incorporated into the upper soil layer of the depression (Ap horizon). The
amount of translocated soil was equivalent to tillage erosion of a decennial
time horizon (Sommer et al., 2016). The change in SOC for each chamber was
monitored by three topsoil inventories, carried out (I) prior to soil
manipulation during April 2009, (II) after soil manipulation during April
2011 and (III) during December 2014. ΔSOC derived through soil
resampling and AC-based C budgets (to determine NECB) was compared for the period
between April 2011 and December 2014 (Fig. 1).
Records of meteorological conditions (1 min frequency) include measurements
of air temperature at 20 and 200 cm height, PAR (photosynthetic active radiation; inside and outside the
chamber), air humidity, precipitation, air pressure, wind speed and
direction. Soil temperatures at depths of 2, 5, 10 and 50 cm were
recorded using thermocouples installed next to the climate station (107,
Campbell Scientific, UT, USA).
The groundwater level (GWL) was measured using tensiometers assuming
hydrostatic equilibrium. The tensiometers were installed at a soil depth of
160 cm at soil profile locations near chamber B and between chambers C and D. The average GWL of both profiles was used for further data
analysis. Data gaps < 2 days were filled using simple linear
interpolation. Larger gaps in GWL did not occur. The measurement site was
cultivated with five different crops during the study period, following a
practice-orientated and erosion-expedited farming procedure. The crop
rotation was silage maize (Zea mays) – winter fodder rye (Secale cereale) – sorghum-Sudan grass
hybrid (Sorghum bicolor x sudanese) – winter triticale (Triticosecale) –
alfalfa (Medicago sativa). Cultivation and fertilization
details are presented in Table A1. Aboveground biomass (NPPshoot)
development was monitored using up to four biomass sampling campaigns during
the growing season, covering the main growth stages. Additional measurements
of leaf area index (LAI) started in 2013. Collected biomass samples were
chopped and dried to a constant weight (48 h at 105 ∘C). The C, N,
K and P contents were determined using elementary analysis (C, N; TruSpec
CNS analyzer, LECO Ltd., Mönchengladbach, Germany) and Kjehldahl
digestion (P, K; AT200, Beckman Coulter (Olympus), Krefeld, Germany and
AAS-iCE3300, Thermo Fisher Scientific GmbH, Darmstadt, Germany). To assess
the potential impact of chamber placement on plant growth, chemical analyses
were carried out for the final harvests of each chamber and were compared to
biomass samples collected next to each chamber.
C budget method
Automatic chamber system
Automatic flow-through non-steady-state (FT-NSS) chamber measurements
(Livingston and Hutchinson, 1995) of CO2 exchange were conducted from
January 2010 until December 2014. The AC system consists of four identical,
rectangular, transparent polycarbonate chambers (thickness of 2 mm, light
transmission ∼ 70 %). Each chamber has a height of 2.5 m and covers a surface area of 2.25 m2 (volume: 5.625 m3). To
adapt for plant height (alfalfa), the chamber volume was reduced to 3.375 m3 in autumn 2013. Airtight closure during measurements was ensured
by a rubber belt that sealed at the bottom of each chamber. A 30 cm
open-ended tube on the slightly concave top of the chambers guided rain
water into the chamber and additionally assured pressure equalization. Two
small axial fans (5.61 m3 min-1) were used for mixing the chamber
headspace. The chambers were mounted onto steel frames with a height of 6 m and lifted between measurements using electrical winches at the top. For
controlling the AC system and data collection, a CR1000 data logger was used
(Campbell Scientific, UT, USA). The CO2 concentration changes over time
were measured within each chamber using a carbon dioxide probe (GMP343,
Vaisala, Vantaa, Finland) connected to a vacuum pump (0.001 m3 min-1; DC12/16FK, Fürgut, Tannheim, Germany). All CO2 probes
were calibrated prior to installation using ±0.5 % accurate gases
containing 0, 200, 370, 600, 1000 and 4000 ppm CO2.
The operation schedule of the AC system, decisively influenced by
agricultural treatments, is presented in Table A1. The chambers closed in
parallel at an hourly frequency, providing one flux measurement per chamber
and hour. The measurement duration was 5–20 min, depending on season and
time of day. Nighttime measurements usually lasted 10 min during the growing
season and 20 min during the non-growing season (due to lower concentration
increments). The length of the daytime measurements was up to 10 min,
depending on low PAR fluctuations (< 20 %). CO2
concentrations (inside the chamber) and general environmental conditions,
such as PAR (SKP215, Skye, Llandrindod Wells, UK) and air temperatures (107,
Campbell Scientific, UT, USA), were recorded inside and outside the chambers
at a 1 min frequency from 2010 to 2012 and a 15 s frequency from October
2012.
CO<inline-formula><mml:math id="M54" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> flux calculation and gap filling
An adaptation of the modular R program script, described in detail by
Hoffmann et al. (2015), was used for stepwise data processing. The
atmospheric sign convention was used for the components of gaseous C
exchange (ecosystem respiration (Reco; sum of autotrophic and
heterotrophic respiration), gross primary production (GPP) and NEE), whereas
positive values for NECB indicate a gain and negative values a loss in SOC.
Based on records of environmental variables and CO2 concentration
change within the chamber headspace, CO2 fluxes were calculated and
parameterized for Reco and GPP within an integrative step.
Subsequently, Reco, GPP and NEE were modeled for the entire
measurement period using climate station data. Statistical analyses, model
calibration and comprehensive error prediction were provided for all steps
of the modeling process.
CO2 fluxes (F, µmol C m-2 s-1) were calculated according
to the ideal gas law (Eq. 1).
F=pVRTA⋅ΔcΔt,
where Δc/Δt is the concentration change over measurement
time, A and V denote the basal area and chamber volume, respectively, and
T
and p represent the air temperature inside the chamber (K) and air pressure.
Because plants below the chambers accounted for < 0.2 % of the
total chamber volume, a static chamber volume was assumed. R is a constant
(8.3143 m3 Pa K-1 mol-1). To calculate Δc/Δt,
data subsets based on a variable moving window with a minimum length of 4 min were used (Hoffmann et al., 2015). Δc/Δt was
computed by applying a linear regression to each data subset, relating
changes in chamber headspace CO2 concentration to measurement time
(Leiber-Sauheitl et al., 2013; Leifeld et al., 2014; Pohl et al., 2015). In
the case of the 15 s measurement frequency, a death band of 5 % was
applied prior to the moving window algorithm. Thus, data noise that
originated from either turbulence or pressure fluctuation caused by chamber
deployment or from increasing saturation and canopy microclimate effects was
excluded (Davidson et al., 2002; Kutzbach et al., 2007; Langensiepen et al.,
2012). Due to the low measurement frequency, no data points were discarded
for records with 1 min measurement frequency (2010–2012). The resulting
CO2 fluxes per measurement (based on the moving window data subsets)
were further evaluated according to the following exclusion criteria: (i)
range of within-chamber air temperature not larger than ±1.5 K
(Reco and NEE fluxes) and a PAR deviation (NEE fluxes only) not larger
than ±20 % of the average to ensure stable environmental
conditions within the chamber throughout the measurement; (ii) significant
regression slope (p≤0.1, t test); and (iii) non-significant tests
(p > 0.1) for normality (Lilliefors adaption of
the Kolmogorov–Smirnov test), homoscedasticity (Breusch–Pagan test) and
linearity of CO2 concentration data. Calculated CO2 fluxes that
did not meet all exclusion criteria were discarded. In cases where more than
one flux per measurement met all exclusion criteria, the CO2 flux with
the steepest slope was chosen.
To account for measurement gaps and to obtain cumulative NEE values,
empirical models were derived based on nighttime Reco and daytime NEE
measurements following Hoffmann et al. (2015). For Reco,
temperature-dependent Arrhenius-type models were used and fitted for
recorded air as well as soil temperatures in different depths (Lloyd and
Taylor, 1994; Eq. 2).
Reco=Rref⋅eE01Tref-T0-1T-T0,
where Reco is the measured ecosystem respiration rate [µmol-1 C m-2 s-1], Rref is the respiration rate at the
reference temperature (283.15 K, Tref), E0 is an activation energy-like parameter, T0 is the starting temperature constant (227.13 K) and
T is the mean air or soil temperature during the flux measurement. Out of the
four Reco models (one model for air temperature; soil temperature
at
2, 5 and 10 cm depth) obtained for nighttime Reco measurements of a
certain period, the model with the lowest Akaike information criterion (AIC)
was used.
GPP fluxes were derived using a PAR-dependent, rectangular hyperbolic light-response function based on the Michaelis–Menten kinetic (Elsgaard et al.,
2012; Hoffmann et al., 2015; Wang et al., 2013; Eq. 3). Because GPP was not
measured directly, GPP fluxes were calculated as the difference between
measured NEE and modeled Reco fluxes.
GPP=GPmax⋅α⋅PARα⋅PAR+GPmax,
where GPP is the calculated gross primary productivity
(µmol-1 CO2 m-2 s-1), GPmax is the maximum rate of C fixation at
infinite PAR (µmol CO2 m-2 s-1), α is the
light use efficiency (mol CO2 mol-1 photons) and PAR is the photon
flux density (inside the chamber) of the photosynthetically active radiation
(µmol-1 photons m-2 s-1). In cases where the
rectangular hyperbolic light-response function did not result in significant
parameter estimates, a non-rectangular hyperbolic light-response function
was used (Gilmanov et al., 2007, 2013; Eq. 4).
GPP=α⋅PAR+GPmax-(α⋅PAR+GPmax)2-4⋅α⋅PAR⋅GPmax⋅θ,
where θ is the convexity coefficient of the light-response equation
(dimensionless).
Due to plant growth and season, parameters of derived Reco and GPP
models may vary with time. To account for this, a moving window
parameterization was performed, by applying fluxes of a variable time window
(2–21 consecutive measurement days) to Eqs. (2)–(4). Temporally overlapping
Reco and GPP model sets were evaluated and discarded in case of
positive (GPP), negative (Reco) or insignificant parameter estimates.
Finally, the model set with the lowest AIC (Reco) was used. If no fit
or a non-significant fit was achieved, averaged flux rates were applied for
Reco and GPP. The length of the averaging period was thereby selected
by choosing the variable moving window with the lowest standard deviation
(SD) of measured fluxes. This procedure was repeated until the whole study
period was parameterized.
Based on continuously monitored temperature and PAR (outside the chamber),
Reco, GPP and NEE were modeled in half-hour steps for the entire study
period. Because GPP was parameterized based on PAR records inside but
modeled with PAR records outside the chamber, no PAR correction in terms of
reduced light transmission was needed. Uncertainty of annual CO2
exchange was quantified using a comprehensive error prediction algorithm
described in detail by Hoffmann et al. (2015).
Modeling aboveground biomass dynamics
Aboveground biomass development (NPPshoot) was predicted using a
logistic empirical model (Yin et al., 2003; Zeide, 1993). From 2010 to 2012,
modeled NPPshoot was based on the relationship between sampling date
and the C content of harvested dry biomass measured during sampling
campaigns (three to four times per year following plant development). For
alfalfa in 2013 and 2014, NPPshoot was modeled based on
measurements of LAI taken once every 2 weeks because no additional biomass sampling was performed
between the multiple cuts per year. To calculate the C content corresponding
to the measured LAI, the relationship between LAI prior to the chamber
harvest and the C content measured in the chamber harvest of all six alfalfa
cuts was used. Daily values of C stored within NPPshoot were calculated
using derived logistic functions.
Calculation of NECB
Annual NECB for each chamber was determined as the sum of annual NEE and
NPPshoot, representing C removal due to the chamber harvest (Eq. 4;
Leifeld et al., 2014). Temporal dynamics in NECB were calculated as the sum
of daily NEE and NPPshoot.
NECBn=∑i=1n[NEEi+CH4+(NPPshooti-Cimport)+ΔDOCi+ΔDICi]
Several minor components of Eq. (5) were not considered (see also
Hernandez-Ramirez et al., 2011). First, C import (Cimport) due to
seeding and fertilization, which was close to zero because the measurement
site was fertilized by a surface application of mineral fertilizer
throughout the entire study period, was not considered. Second, methane (CH4-C) emissions,
which were measured manually at the same experimental field but did not
exceed a relevant order of magnitude (-0.01 g C m-2 yr-1) were
not included in the NECB calculation. Third, lateral C fluxes,
originating from dissolved organic carbon (DOC) and dissolved inorganic carbon (DIC) as well
as particulate soil organic carbon (SOCp), were not considered. In addition to the rather
small magnitude of the subsurface lateral C fluxes in soil solution (Rieckh
et al., 2012), it was assumed that their C input equaled C output at the
plot scale. Lateral SOCp transport along the hillslope was excluded by
grassland stripes established between experimental plots in 2010 (Fig. 1 in
Sommer et al., 2016).
Soil resampling method
To obtain ΔSOC using the soil resampling method, soil samples were
collected three times during the study period. Initial SOC along the
topographic gradient was monitored prior to soil manipulation during April
2009 at two soil pits, which were sampled by pedogenetic horizons. After
soil manipulation, a 5 m raster sampling of topsoils (Ap horizons) was
performed during April 2011. Each Ap horizon was separated into an upper
(0–15 cm) and lower segment (15–25 cm), which were analyzed separately for
bulk density, SOC, total nitrogen (Nt) and coarse fraction (< 2 mm) (data not shown).
From these data, SOC and Nt mass densities were calculated separately for
each segment and finally summed up for the entire Ap horizon (0–25 cm). The
mean SOC and Nt content for the Ap horizon of each raster point was
calculated by dividing SOC or Nt mass densities (0–25 cm) through the
fine-earth mass (0–25 cm). In December 2014, composite soil samples of the
Ap horizon were collected. The composite samples consist of samples from
four sampling points in a close proximity around each chamber. Prior to
laboratory analysis, coarse organic material was discarded from collected
soil samples (Schlichting et al., 1995). Thermogravimetric desiccation at
105 ∘C was performed in the laboratory for all samples to
determine bulk densities (Mg m-3). Bulk soil samples were air dried,
gently crushed and sieved (2 mm) to obtain the fine fraction (particle size
< 2 mm). The total carbon and total nitrogen contents were
determined by elementary analysis (TruSpec CNS analyzer, LECO Ltd.,
Mönchengladbach, Germany) using carbon dioxide via infrared detection after
dry combustion at 1250 ∘C (DIN ISO10694, 1996), in duplicate. As
the soil horizons did not contain carbonates, total carbon was equal to SOC.
Time series of CO2 exchange (a–d) for the four chambers of the
AC system during the study period from 2010 to 2014. Reco (black), GPP (light gray) and NEE (dark gray) are shown as daily
sums (y axis). NEEcum is presented as a solid line, representing the
sum of continuously accumulated daily NEE values (secondary y axis). The
presented values display cumulative NEE following soil manipulation to the
end of 2014. Note the different scales of the y axes. The gray shaded area
represents the period prior to soil manipulation. The dashed vertical line
indicates the soil manipulation. Dotted lines represent harvest events.
![]()
Time series of modeled aboveground biomass development
(NPPshoot) (a–d) for the four chambers of the AC system during the
study period from 2010 to 2014. NPPshoot is shown as cumulative values.
The presented values display cumulative NPPshoot following soil
manipulation to the end of 2014. The biomass model is based on biomass
sampling (2010–2012) and LAI measurements taken once every 2 weeks (2013–2014) during crop
growth (gray dots). C removal due to chamber harvests is shown by black
dots. The gray shaded area represents the period prior to soil manipulation.
The dashed vertical line indicates the soil manipulation. Dotted lines
represent harvest events.
![]()
Uncertainty prediction and statistical analysis
Uncertainty prediction for NECB derived by the C budget method was performed
according to Hoffmann et al. (2015), following the law of error propagation.
To test for differences in topsoil SOC (SOCAp) and Nt
stocks in soil resampling performed after soil manipulation in 2010 and
2014, a paired t test was applied. Computation of uncertainty prediction and
calculation of statistical analyses were performed using R 3.2.2.
Discussion
Accuracy and precision of applied methods
Despite the similar magnitude and tendencies of the observed NECB and
ΔSOC values, both methods were subject to numerous sources of
uncertainty, representing the different concepts they are based on (see
introduction). These errors affect the accuracy and precision of observed
NECB and ΔSOC values differently, which might help to explain
differences between the soil resampling and the C budget method.
The soil resampling method is characterized by high measurement precision,
which allows for the detection of relatively small changes in SOC. Related
uncertainty in derived spatial and temporal ΔSOC dynamics is
therefore mainly attributed to the measurement accuracy, affected by
sampling strategy and design (Batjes and van Wesemael, 2015; De Gruijter et
al., 2006). This includes (i) the spatial distribution of collected samples,
(ii) the sampling frequency, (iii) the sampling depth and (iv) whether
different components of soil organic matter (SOM) are excluded prior to
analyses. The first aspect determines the capability of detecting the inherent
spatial differences in SOC stocks. This allows the conclusion that point
measurements do not necessarily represent AC measurements, which integrate
over the spatial variability within their basal area. The second aspect
defines the temporal resolution, even though the soil resampling method is
not able to perfectly separate spatial from temporal variability because
repeated soil samples are biased by inherent spatial variability of the
measurement site. The third aspect sets the vertical system boundary, which
is often limited because only topsoil horizons are sampled within a number
of soil monitoring networks (Van Wesemael et al., 2011) and repeated soil
inventories (Leifeld et al., 2011). Similarly, the fourth aspect defines
which components of SOM are specifically analyzed. Usually, coarse organic
material is discarded prior to analysis (Schlichting et al., 1995) and
therefore total SOC is not assessed (e.g., roots, harvest residues).
Chamber-specific annual sums of CO2 exchange (Reco, GPP,
NEE), NPPshoot, NECB and ΔSOC (±uncertainty), as well as
corresponding environmental variables measured during the study period from
2010 to 2014.
Year
Crop rotation
Position
Reco
GPP
NEE
NECB*
NPPshoot
NPPshoot
SOC to
SOC in
ΔSOC
Nt to
Nt in Ap
Precip.
GWL
Harvested
Modeled
N
P
K
1 m depth
Ap horizon
1 m depth
(g C m-2)
(g C m-2)
(g m-2)
(kg m-2
(kg m-2
(g C m-2)
(kg m-2
(kg m-2
(mm)
(cm)
1 m-1)
0.3 m-1)
1 m-1)
0.3 m-1)
2010
Maize
A (upper)
1014 ± 9
-1845 ± 8
-831a ± 12
86 ± 66
744
745a±65
28.1
5.0
25.6
11.6
5.1
–
1.3
0.6
516
135
B (upper middle)
987 ± 11
-1970a ± 8
-983 ± 13
251 ± 66
727
732a ± 64
24.7
4.1
18.0
9.1
4.2
0.9
0.4
103
C (lower middle)
1064 ± 38
-2000a ± 11
-935a ± 40
190 ± 77
744
745a ± 65
25.5
4.2
16.9
9.1
4.2
0.9
0.4
95
D (lower)
1110 ± 21
-1737 ± 10
-627a ± 23
-118 ± 69
744
745a ± 65
25.0
4.2
18.2
12.8
5.0
1.3
0.5
69
2011
Maize
A (upper)
891 ± 13
-2022 ± 18
-1131a ± 22
-149 ± 103
1238
1280a ± 101
29.5
5.4
30.2
10.5
3.5
–
1.1
0.4
618
129
B (upper middle)
855a ± 10
-1894 ± 13
-1039a ± 16
-169 ± 96
1167
1208a ± 95
36.4
5.9
32.7
8.7
3.4
0.9
0.4
97
C (lower middle)
980 ± 14
-2062 ± 25
-1082 ± 28
-79 ± 95
1115
1161a ± 91
33.7
5.6
32.9
9.0
3.7
0.9
0.4
87
D (lower)
843a ± 31
-1730 ± 8
-888 ± 32
-59 ± 80
900
947a ± 73
35.0
5.7
31.8
12.2
4.0
1.3
0.4
61
2012
Winter wheat
A (upper)
1058 ± 86
-2659 ± 12
-1600 ± 87
648 ± 104
297**/634
952a ± 56
36.3
6.3
42.6
–
–
585
139
B (upper middle)
1075 ± 8
-2591 ± 11
-1516 ± 13
472 ± 65
310**/727
1044a ± 64
33.3
5.8
37.5
107
Sorghum
C (lower middle)
1286 ± 8
-2617 ± 9
-1331 ± 12
346 ± 60
310**/665
985a ± 59
32.7
5.4
35.5
87
D (lower)
1044 ± 10
-2194 ± 9
-1150 ± 13
430 ± 39
299**/420
720a ± 37
33.9
5.8
40.4
61
2013
Alfalfa
A (upper)
1140 ± 83
-1583 ± 9
-443 ± 83
43 ± 91
290
400a,b ± 37
14.0
1.7
11.6
499
154
B (upper middle)
1283 ± 80
-1819 ± 8
-536 ± 80
93 ± 86
304
443b ± 32
14.7
1.8
12.1
122
C (lower middle)
1438 ± 20
-1726 ± 7
-288 ± 22
-107 ± 36
324
395a± 29
15.6
1.9
12.9
94
D (lower)
1587 ± 80
-2036 ± 8
-448 ± 80
6 ± 87
329
442b ± 34
15.9
2.0
13.2
68
2014
A (upper)
1161 ± 15
-1615 ± 7
-455a ± 16
-126 ± 26
605
581a ± 20
29.2
3.6
24.2
10.9
3.9
376
1.2
0.5
591
181
B (upper middle)
1443 ± 18
-2063 ± 7
-619a ± 19
52 ± 28
635
567a ± 20
30.7
3.8
25.4
8.9
3.5
156
0.9
0.4
149
C (lower middle)
1683 ± 18
-2111 ± 6
-428 ± 19
-36 ± 26
632
535a ± 18
30.5
3.8
25.3
9.0
3.7
0
0.9
0.5
121
D (lower)
1584 ± 12
-2113 ± 14
-528 ± 19
-52 ± 28
587
580a ± 21
28.3
3.5
23.5
12.5
4.2
276
1.3
0.4
95
Annual average
A (upper)
1063 ± 49
-1970 ± 12
-901 ± 52
98 ± 43
766
803 ± 54
27.3
4.3
27.2
–
94 ± 43
–
573
151
(2011–2014)
B (upper middle)
1164 ± 29
-2092 ± 10
-919 ± 32
104 ± 37
786
815 ± 53
28.8
4.3
26.9
39 ± 43
119
C (lower middle)
1347 ± 15
-2129 ± 12
-779 ± 20
10 ± 30
762
769 ± 49
28.1
4.2
26.7
0 ± 46
97
D (lower)
1265 ± 33
-2018 ± 10
-739 ± 38
67 ± 32
634
672 ± 41
28.3
4.3
27.2
69 ± 47
71
Site
1209 ± 32
-2052 ± 11
-843 ± 36
78 ± 18
737
765 ± 49
28.1
4.3
27.0
51 ± 18
156
1 For comparability reasons the NECB is given using
the soil sign convention (negative values = soil C loss;
positive values = soil C gain).2 NPPshoot is based on biomass samples collected next to
each chamber because no chamber harvest was performed for winter fodder rye in 2012;
superscript letters indicate non-significant differences (Wilcoxon rank-sum
test; p value > 0.05) between measured CO2 fluxes and
NPPshoot.
Temporal and spatial dynamics in cumulative NECB and ΔSOC
throughout the study period based on (a) the C budget method
(measured–modeled, black lines) and (b) the soil resampling method (linear
interpolation, gray lines), respectively. The gray shaded area represents
the period prior to soil manipulation. The dashed vertical line indicates
the soil manipulation. Dotted lines represent harvest events. Temporal
dynamics in NECB revealed by the C budget method allow for the
identification of periods that are most important for changes in SOC. Major
spatial deviation occurred during the maximum plant growth period (May to
September). The proportion (%) of these periods with respect to the
standard deviation of estimated annual NECB accounted for up to 79 %.
![]()
Average annual ΔSOC observed after soil manipulation (April
2011 to December 2014) by soil resampling and the C budget method for (a)
the entire measurement site and (b) single chamber positions within the
measured transect. ΔSOC represents the change in carbon storage,
with positive values indicating C sequestration and negative values
indicating C losses. Error bars display estimated uncertainty for the C
budget method and the analytical error of ±5 % for the soil
resampling method. A performed Wilcoxon rank-sum test showed no significant
difference between NECB and ΔSOC values obtained by both
methodological approaches for all four chambers (p value = 0.25).
![]()
In comparison, the C budget method considers any type of organic material
present in soil by integrating over the total soil depth. As a result, both
methods have a different validity range and area, which makes direct
quantitative comparison more difficult. This may explain the higher uptake
reported for three out of four chamber positions in the case of the C budget
method.
In contrast to the soil resampling method, we postulate a higher accuracy
and a lower precision in the case of the AC-based C budget method. The
reasons for this include a number of potential errors affecting especially
the measurement precision of the AC system, whereas over a constant area and
maximum soil depth, integrated AC measurements increase measurement
accuracy. First, it is currently not clear whether microclimatological and
ecophysiological disturbances due to chamber deployment, such as the
alteration of temperature, humidity, pressure, radiation and gas
concentration, may result in biased C flux rate estimates (Juszczak et al.,
2013; Kutzbach et al., 2007; Lai et al., 2012; Langensiepen et al., 2012).
Second, uncertainties related to performed flux separation and gap-filling
procedures may influence the obtained annual gaseous C exchange
(Gomez-Casanovas et al., 2013; Görres et al., 2014; Moffat et al., 2007;
Reichstein et al., 2005). Although continuous operation of the AC system
should allow for direct derivation of C budgets from measured CO2
exchange and annual yields, in practice, data gaps always occur. To fill the
measurement gaps, temperature- and PAR-dependent models are derived and used
to calculate Reco and GPP, respectively (Hoffmann et al., 2015). Due to
the transparent chambers used, modeled Reco is solely based on
nighttime measurements. Hence, systematic differences between nighttime and
daytime Reco will yield an over- or underestimation of modeled
Reco. Because modeled Reco is used to calculate GPP fluxes, GPP
will be affected in a similar manner. However, the systematic over- or
underestimation of fluxes in both directions may counterbalance the computed
NEE, and estimated C budgets may be unaffected. Third, the development of
NPPshoot underneath the chamber might be influenced by the permanently
installed AC system. Fourth, several minor components such as leaching
losses of DIC and DOC, C transport
via runoff and atmospheric C deposition were not considered within the
applied budgeting approach (see also Sect. 2.7).
Despite the uncertainties mentioned above, error estimates for annual NEE in
this study are within the range of errors presented for annual NEE estimates
derived from EC measurements (30 to 50 g C m-2 yr-1) (e.g.,
Baldocchi, 2003; Dobermann et al., 2006; Hollinger et al., 2005) and below
the minimum detectable difference reported for most repeated soil
inventories (e.g., Batjes and Van Wesemael, 2015; Knebl et al., 2015;
Necpálová et al., 2014; Saby et al., 2008; Schrumpf et al., 2011;
VandenBygaart, 2006).
Plausibility of observed <inline-formula><mml:math id="M462" display="inline"><mml:mi mathvariant="normal">Δ</mml:mi></mml:math></inline-formula>SOC
Both the soil resampling and the C budget method showed C gains during the
4 years following soil manipulation. A number of authors calculated
additional C sequestration due to soil erosion (Berhe et al., 2007; Dymond,
2010; VandenBygaart et al., 2015; Yoo et al., 2005), which was explained by
the burial of replaced C at depositional sites and dynamic replacement at
eroded sites (e.g., Doetterl et al., 2016). This is in accordance with
erosion-induced C sequestration postulated by Berhe and Kleber (2013)
and Van Oost et al. (2007), for example. In addition, observed C sequestration could also
be a result of the manipulation-induced saturation deficit in SOC. By adding
topsoil material from an eroded unsaturated hillslope soil, the capacity
and efficiency of sequestering C was theoretically increased (Stewart et al.,
2007). Hence, additional C was stored at the measurement site. This might be
due to physicochemical processes, such as physical protection in macro- and
microaggregates (Six et al., 2002) or chemical stabilization by clay and
iron minerals (Kleber et al., 2015).
Irrespective of the similar C gain observed by both methods, crop-dependent
differences in NECB and thus ΔSOC were only revealed by the C budget
method. The reason is the higher temporal resolution of AC-derived C
budgets, displaying daily C losses and gains. Observed crop-dependent
differences in NECB are in accordance with Kutsch et al. (2010), Jans
et al. (2010), Hollinger et al. (2005) and Verma et al. (2005), for example, who reported
comparable EC-derived C balances for, inter alia, maize, sorghum and alfalfa.
In 2012, substantial positive annual NECB values were observed. Due to low
precipitation during May and June, germination and plant growth of
sorghum-Sudan grass was delayed (Fig. 4). As a result, the reproductive
phenological stage was drastically shortened. This reduced C losses prior to
harvest due to higher Reco : GPP ratios (Wagle et al., 2015). In
addition, the presence of cover crops during spring and autumn could have
increased SOC, as reported by Lal et al. (2004), Ghimire et al. (2014) and
Sainju et al. (2002). No additional C sequestration was observed for alfalfa
in 2013 and 2014 or for the lower middle chamber position, which acted
neither as a net C source nor sink (Table 1, Fig. 5). This opposes the
assumption of increased C sequestration by perennial grasses (Paustian et
al., 1997) or perennial crops (Zan et al., 2001). However, NEE estimates of
alfalfa were within the range of -100 to -400 g C m-2, which is typical
for forage crops (Lolium, alfalfa, etc.) in different agro-ecosystems (Bolinder et
al., 2012; Byrne et al., 2005; Gilmanov et al., 2013; Zan et al., 2001). In
addition, Alberti et al. (2010) reported a soil C loss of > 170 g C m-2 after crop conversion from continuous maize to alfalfa,
concluding that no effective C sequestration occurs in the short term.
Regardless of the crop type, the AC-derived dynamic NECB values showed that
up to 79 % of the standard deviation of estimated annual NECB occurred
during the growing season and the main plant growth period from the
beginning of July to the end of September.