Biogeosciences Discuss., 6, 9977–10005, 2009
www.biogeosciences-discuss.net/6/9977/2009/
© Author(s) 2009. This work is distributed under
the Creative Commons Attribution 3.0 License.
Biogeosciences
Discussions
Biogeosciences Discussions is the access reviewed discussion forum of Biogeosciences
Small scale spatial heterogeneity of soil
respiration in an old growth temperate
deciduous forest
A. Jordan1,2, G. Jurasinski2, and S. Glatzel2
1Georg-August-University Goettingen, Department of Geography, Landscape Ecology-Unit,
Goldschmidtstrasse 5, 37077 Go¨ttingen, Germany
2University of Rostock, Faculty of Agricultural and Environmental Sciences, Landscape
Ecology and Site Evaluation, Justus-von-Liebig-Weg 6, 18059 Rostock, Germany
Received: 17 August 2009 – Accepted: 29 September 2009 – Published: 22 October 2009
Correspondence to: A. Jordan (albrecht.jordan@gmx.de)
Published by Copernicus Publications on behalf of the European Geosciences Union.
9977
Abstract
The large scale spatial heterogeneity of soil respiration caused by differences in site
conditions is quite well understood. However, comparably little is known about the mi-
cro scale heterogeneity within forest ecosystems on homogeneous soils. Forest age,
soil texture, topographic position, micro topography and stand structure may influence5
soil respiration considerably within short distance. In the present study within site spa-
tial heterogeneity of soil respiration has been evaluated. To do so, an improvement of
available techniques for interpolating soil respiration data via kriging was undertaken.
Soil respiration was measured with closed chambers biweekly from April 2005 to
April 2006 using a nested design (a set of stratified random plots, supplemented by 210
small and 2 large nested groupings) in an unmanaged, beech dominated old growth
forest in Central Germany (Hainich, Thuringia). A second exclusive randomized design
was established in August 2005 and continually sampled biweekly until July 2007.
The average soil respiration values from the random plots were standardized by
modeling soil respiration data at defined soil temperature and soil moisture values. By15
comparing sampling points as well as by comparing kriging results based on various
sampling point densities, we found that the exclusion of local outliers was of great
importance for the reliability of the estimated fluxes. Most of this information would have
been missed without the nested groupings. The extrapolation results slightly improved
when additional parameters like soil temperature and soil moisture were included in20
the extrapolation procedure. Semivariograms solely calculated from soil respiration
data show a broad variety of autocorrelation distances (ranges) from a few centimeters
up to a few tens of meters.
The combination of randomly distributed plots with nested groupings plus the in-
clusion of additional relevant parameters like soil temperature and soil moisture data25
permits an improved estimation of the range of soil respiration, which is a prerequisite
for reliable interpolated maps of soil respiration.
9978
1 Introduction
Soil CO2 efflux is one of the largest carbon flux components. Globally it accounts
for 20–40% (50–75GtC/yr) of the total annual input of carbon dioxide into the atmo-
sphere (Houghton and Woodwell, 1989; Raich and Schlesinger, 1992; Schimel, 1995;
Lankreijer et al., 2003). Even small shifts between carbon release and photosynthetic5
uptake can have a significant influence on the atmospheric carbon budget (Pregitzer
and Euskirchen, 2004).
4.1 bill hectares of land on earth are covered with boreal, temperate and tropical
forests containing up to 80% of the terrestrial aboveground carbon and 40% of total
soil C (Dixon et al., 1994; Pregitzer and Euskirchen, 2004). However, mean soil efflux10
rates do not only vary between the major biome types, they also show high variabil-
ity within the same biome and even within the same measurement site (Raich and
Schlesinger, 1992). Thus, there are high uncertainties in the total annual values of site
scale soil efflux rates (a site is understood here as a particular ecosystem e.g., a forest
or a wetland). Small-scale heterogeneities in soil efflux rates are, inter alia, caused by15
the high variability of soil structure (Bouma and Bryla, 2000), soil moisture (Davidson
et al., 1998), bacterial and fungal communities (Gomoryova, 2004), root density (Han-
son et al., 2000; Soe and Buchmann, 2005), soil organic matter content, wind speed
at the soil surface and pressure patterns (Lankreijer et al., 2003; Martin and Bolstad,
2005).20
Generally, the term “spatial heterogeneity” is used to refer to differences between
measurement sites. Most of these sites are selected just because of their different
sample characteristics, for example their different soil or vegetation type, stand struc-
ture or stand age (Law et al., 2001; Salimon et al., 2004; Vogel et al., 2005). Studies
describing within site “spatial heterogeneity”, that means micro scale spatial hetero-25
geneity within old growth forests with otherwise relatively homogeneous site conditions
are extremely rare. Many studies in the past few years partitioned soil respiration into
autotrophic and heterotrophic components (Kuzyakov, 2006; Subke et al., 2006) and
9979
advances have been made in understanding the dynamics of a wide range of soil C
pools and their effects on soil respiration (Baggs, 2006; Kalbitz et al., 2000; Ryan and
Law, 2005). However, it still remains unclear on which scales factors influencing soil
respiration vary, while influencing total soil respiration rate.
Having said this, it is very likely that these factors do not take effect on the same5
scales within a given site. Thus, an approach spanning various spatial scales seems
to be adequate. Therefore, in the present study we adopted a three-fold strategy. First,
we set up nested groups of sampling plots. Second, we recorded data from stratified
randomly distributed plots. Third, additional information gathered at the sampling points
was used via kriging to improve the reliability of the site scale estimates of soil efflux10
rates. With this approach we aimed for a better understanding of the spatial variability
of soil respiration in old growth temperate deciduous forests. Specifically we tested the
following hypotheses:
1) The environmental controls for soil respiration vary on different spatial scales
within forest sites. Therefore, a multiple scale sampling approach improves the15
reliability of site scale soil efflux rate estimates extrapolated from point measure-
ments.
2) The consideration of additional parameters with control potential for soil respira-
tion further improves the extrapolation results.
2 Material and methods20
2.1 Site
The study site is located in the core zone of the Hainich National Park in Thuringia,
Central Germany (51◦ 04′ 46′′N, 10◦ 27′ 08′′ E, 440ma.s.l.). Sampling locations were
established within the fetch area of an Eddy Covariance Tower that is run by the MPI
for Biogeochemistry Jena, Germany. The mean annual temperature and precipitation25
9980
are 6–7◦C and 750mm. Soils are chromic cambisols (ISSS, 1998) on limestone. Veg-
etation structure is that of an old growth forest with a large amount of dead wood and
high species diversity. The oldest trees reach 250 yr of age. The forest is dominated
by Fagus sylvatica L. Additionally Fraxinus excelsior L., Acer pseudoplatanus L., and
Acer platanoides L. occur in smaller fractions.5
2.2 Sampling design
We implemented two different sampling approaches. On the one hand we followed
a double nested, stratified random approach (nested series) to determine the maximum
range of autocorrelation as a prerequisite for extrapolation (Fig. 1). 79 measurement
locations (called “plots” in the following) were arranged in this series. We superimposed10
a grid on a map of the site that divides it into 43 rectangular cells with an edge length of
34m. In each of the grid cells one plot was placed in a random location to reduce the
maximum distance between the points and to avoid large gaps between them. Thus,
43 plots were distributed in a stratified random layout across the whole study site. To
obtain spatially nested plots 2 cells were chosen randomly from the grid and 36 plots15
were placed in 4 groups within them resulting in two small and two larger plot groupings
nested within each other (see Fig. 1 for details). This double nested approach allowed
computing soil respiration data with three different spatial resolutions. Additionally 81
plots were located completely random across the study site (random series).
All plots were equipped with PVC collars with an inner diameter of 10 cm and a height20
of 7 cm. They were inserted 3 cm into the soil two weeks before the measurements
started to avoid disturbance. Each collar was laterally fixed with three tent pegs to
secure fix them against removal by animals.
2.3 Sampling procedure
Both series were sampled biweekly. For the nested series sampling was carried out25
from April 2005 to April 2006. The random series was sampled from August 2005 to
9981
June 2007. During the time when both series were sampled (August 2005 to April
2006) the biweekly sampling was carried out alternately.
Soil respiration was recorded with a portable infrared gas analyser (IRGA, EGM-4)
connected to a closed soil respiration chamber (SRC-1), both devices by PP Systems,
Hitchin, UK. The soil respiration rate was calculated as the rate of change in CO25
concentration within the chamber headspace. The EGM-4 allows for an automatic
calculation of emission rates provided that the volume of the chamber and the surface
area of the soil are set. The CO2 concentration was monitored each 4.8 s over an
interval of 124 s. Only the final results were saved and recorded.
In preparation of both measurement series, a set of comparative measurements was10
performed using a second soil respiration measurement device (Li-cor 6400-09, Li-
cor, Inc., Lincoln, Nebraska, USA), to ensure the quality of the recorded data. Due
to the high reproducibility and the stability of both measurement results, the EGM-4
measurements were seen as reliable and sufficient.
In addition to the recording of respiration rates soil temperature was measured next15
to each plot at a depth of 0–10 cm below the litter layer by digital precision thermome-
ters (Greisinger, Regenstauf, Germany for the nested series; Testo, Lenzkirch, Ger-
many for the random series). Due to the year round measurement campaigns, a few
data gaps occurred at ground frost events. These were closed with data from the
continuously measuring soil sensors of the Max-Planck-Institute for Biogeochemistry,20
Jena. These sensors collect data at depths of −2, −5, −15, −30 and −50 cm. To
assure comparability, the average temperature of the sensors in the depths of −2 and
−5 cm were calculated. These values showed a very close and significant correlation
with the average value of the own measurements at non-ground frost days.
From the 14 April 2005 until the 12 April 2006 soil moisture was determined gravimet-25
rically. During each measurement event, gravimetric soil water content was determined
next to each plot. Therefore soil samples were taken twice at each plot with a small
auger up to a depth of −15 cm. These wet samples were weighted, oven-dried for 24 h
at 105◦C to stable weight and weighted again. The soil moisture was calculated as the
9982
percentage difference between wet and dry weight. From April 2006 until the end of
the measurements soil moisture was determined with frequency domain reflectrometry
(fdr) at 0–6 cm soil depth below the litter layer with the ThetaProbe Soil Moisture Sen-
sor – ML-2x and the Soil Moisture Meter – HH2 (AT-Delta-T Devices Ltd., Cambridge,
UK). In contrast to soil temperature, data gaps during ground frost periods could not5
be closed using data from the eddy covariance tower, because of the large spatial
variability of soil moisture measurements at single measurement dates.
2.4 Statistics
Tests for normal distribution with the Kolmogoroff-Smirnoff testing algorithm showed
that non-parametric testing procedures were necessary, so we proceeded with Spear-10
man rank correlations. All statistical calculations were carried out with STATISTICA 7.0
(Statsoft, Oklahoma, USA).
To estimate the dependency of seasonal variation of soil respiration from soil tem-
perature, the temperature functions were fitted by regression to the simple exponential
equation:15
R = keaT (1)
with R=soil respiration [ gCO2m
−2 a−1], T=soil temperature (0–10 cm) [◦C], k and a are
constants (Davidson et al., 1998; Knohl et al., 2008). Q10 represents the temperature
response of soil respiration and is calculated with the simple formula:
Q10 = e
a10 (2)20
Saiz et al. (2006) and Stoyan et al. (2000) found that soil moisture is more impor-
tant for the determination of spatial effects in temperate deciduous forests than soil
temperature. Therefore, we consider soil moisture in the model (Eq. 3).
R = keaTebθ, or ln(R) = ln(k) + a(t) + b(θ) (3)
9983
with θ=soil moisture measured at 0–6 cm [ vol%] and a, b are fitted constant. The
exponential equation taken from Knohl et al. (2008), was log-transformed to a linear
model in order to conduct linear regression for the estimation of the parameters. Soil
respiration rates were standardized using soil moisture and soil temperature. Eq. 4
presents the modified version of Eq. (3), where RδTδθ represents the modelled soil res-5
piration for each collar, T and θ are the standardized soil temperature and soil moisture.
δT and δθ are the differences between standardized and measured temperature and
moisture values and k, a and b are constants again.
RδTδθ = ke
a(T+δT )eb(θ+δθ) (4)
= keaTeaδTebθebδθ10
= ReaδTebδθ
Rstandard =
RδTδθ
eaδTebδθ
(5)
Equation (4) can be transformed to separate R, so R is replaced by Rstandard (Eq. 5).
The modelled RδTδθ term in Eq. (5) can be substituted by the measured soil respiration
values. We used the trend identification tool of ArcGIS 9.2 to identify spatial trends15
by plotting single soil respiration measurement events within a three dimensional dia-
gram. To estimate the confidence of the observed total mean flux, we calculated N, the
number of measurement locations needed for a given precision level (Davidson et al.,
2002; Yim et al., 2003; Adachi et al., 2005; Knohl et al., 2008) using Eq. (6):
N =
[
tα×s
range/2
]2
, (6)20
with t the Student’s t-statistic, s the standard deviation of one set of daily measure-
ments, and range the desired interval around the daily mean. N was calculated for
each measurement date separately. N was averaged to obtain a valid N for the entire
study.
9984
2.5 Geostatistical analysis
We extrapolated the pointwise measurements using ordinary kriging. Kriging is based
on the construction of experimental semivariograms from the data (Goovaerts, 1999;
Webster and Oliver, 2001). To optimize the kriging results we varied the lag distance
until we obtained the best possible parameter estimates. Kriging was carried out using5
the Geostatistical Analyst extension of ArcGIS 9.2 (ESRI, Redlands, USA).
We used all sampling points located within the two grid cells with the nested group-
ings to quantify autocorrelation at different spatial scale. Thus, we used three different
sets of plots for estimating the kriging parameters. First, only the stratified random plots
of the nested series with relatively large inter-plot distances were used. Then, the data10
of the larger groupings from the nested series and finally the data of the small scale
groupings were added. The maps as well as the parameter estimates resulting from
ordinary kriging with these three datasets were compared (Fig. 6). The cell size of the
maps was set to 10×10 cm and the maps were classified in 0.01 gCO2m−2 h−1 units,
the same resolution as the applied measurement equipment. “Hawth’s Analysis Tool”15
(http://www.spatialecology.com, extension to ArcGIS) was used to count the number of
equal values per class. Thus, average soil respiration rates were calculated for different
sampling densities.
To estimate annual fluxes based on the random series with 81 data points, regres-
sion equations using measured soil respiration, soil moisture and soil temperature data20
were calculated for each plot. Soil temperature values measured at the eddy covari-
ance facility were used to estimate the mean flux for each plot each day of the year.
Two methods were used for further calculation: On the on hand we calculated the final
annual flux of the site from the average sum of all plots, on the other hand we used
a geostatistical approach. Therefore average standardised soil respiration values of25
each plot were extrapolated using ordinary kriging. A minimum convex polygon was
created around the measurement points and reduced by 10m inwards to avoid bor-
der effects. The resulting polygon was used to clip the ordinary kriging result. Zonal
9985
statistics were performed to calculate the total soil respiration rate at the study site.
3 Results
3.1 Seasonal variation of soil respiration
Looking solely at the flux rates of the random series the measurements started
with average soil respiration rates of 0.48 gCO2×m−2×h−1, with a soil temperature5
of 14◦C and a soil moisture of 40% (Fig. 2). From September 2005 the average
soil respiration rate decreased constantly until it reached an absolute minimum of
0.04 gCO2×m−2×h−1 in January 2006. Then, in spring 2006, the soil temperature
increased rapidly. This was followed with a little delay by the soil respiration rate. The
two lower respiration values in the middle of May and July 2006 seem to correspond10
with temporal decreases in soil moisture availability. Maximum soil temperature and
minimum soil moisture was measured on 26 July 2006.
The seasonal variation of soil respiration is clearly temperature dependent (Fig. 3,
r2=0.69, Q10=2.3). Q10 amounts to 3.9. The total average soil respiration value is
0.28 gCO2m
−2 h−1. However, the relation between soil respiration and soil tempera-15
ture was much closer at collars with lower soil moisture content (0–30%Vol., r2=0.69,
Q10=2.3) compared to wetter plots (>60–90%Vol., r
2=0.32, Q10=4.1). The highest
average respiration rates (0.30 gCO2m
−2 h−1) were observed at medium soil moisture
between >30–60 [%Vol.].
Soil temperature did not vary much in space (average soil temperature variability at20
a given measurement date was 1.3◦C), whereas soil moisture measurements showed
a considerable spatial heterogeneity as well as high seasonal variability. Total average
volumetric water content was 37% with a wide range from 5% to 88%. A multiple
regression model including soil moisture and soil temperature as explanatory variables
slightly improves the explanation (R2 increased from 0.60 to 0.63). This model was25
used for the regionalization.
9986
3.2 Spatial distribution
3.2.1 Spatial trends
When averaged in time, soil respiration does not show a spatial trend, because the
presence and absence of soil respiration hotspots does not follow a regular pattern.
Therefore, ordinary kriging can be carried out without problems.5
3.2.2 Number of measurement plots required
There are few but highly significant outliers regarding annual soil respiration rates even
within short distances. This underlines the need to calculate the number of measure-
ment points required for different levels of accuracy beforehand. Measurements from
6±4 locations are necessary for an 80% confidence level with an error limit of 30%10
around the full population mean (Table 1). Higher levels of precision require a higher
number of measurement locations. At a 95% confidence level, 16±11 collars are nec-
essary for the same large error limit. Reducing this limit to 20% precision increases
the required number of collars up to 35±26, further precision can be achieved by using
140±103 at an error limit of 10%.15
3.2.3 Kriging preparation
Data point density was varied within two grid cells to infer the appropriate parameters
for kriging (Fig. 4). Whereas the calculated average soil respiration rates in one of the
grid cells changed little at different sampling densities, soil respiration rates decreased
significantly with higher numbers of collars within the other. Both cells are character-20
ized by artificial block structures at low and medium sampling point density, indicating
a distance between the sampling points wider than the range of the semivariograms.
Much of the block structures can be smoothed by using a coarser scale for soil respi-
ration, but not all.
9987
3.2.4 Semivariograms
Varying lag distances when calculating semivariograms resulted in markedly heteroge-
neous autocorrelation ranges. When using a lag distance of 10m the resulting semi-
variograms reveal a lack of spatial autocorrelation (Fig. 5a, b). Zooming in by choosing
smaller lag sizes (Fig. 5c,d) whilst reducing the dataset by eliminating outliers, results5
in small range autocorrelation. Furthermore, ranges vary considerably from date to
date. However, the average range of autocorrelation was determined at 9m, using lag
sizes of 2m. This lag size was used in all further steps of analysis.
3.2.5 Maps of soil respiration
Maps of soil respiration rates were extrapolated based on ordinary kriging. On the10
one hand data from the nested series was used (Fig. 6a). The resulting map exhibits
a high level of uncertainty due to the short range of the calculated semivariogram
(<20m) and the low sampling point density within the study site with distances between
adjacent collars >25m (except within the observational windows). Even at class sizes
of 0.03 gCO2m
−2 h−1 the map has many block structures. The corresponding map of15
predicted standard errors shows concentric rings around single measurement collars
with low standard error and great areas with high standard errors outside the range
of the calculated semivariogram (Fig. 6b). The negative mean error indicates that the
interpolated soil respiration data are slightly overestimated by the model. On the other
hand data from the random series was used for interpolation (Fig. 6c). Due to the20
higher measurement point density, the predicted standard error map is much smoother
compared to results based on the nested series data (Fig. 6d).
A further interpolation run is based on random series and standardized soil respira-
tion rates (Fig. 6e). To achieve standardized soil respiration rates average soil respira-
tion was used with Eq. (5). For the 2006 series it amounts to 0.26 gCO2m
−2 h−1 and25
a typical soil temperature of 11.6◦C results. This is much higher than the total average
soil temperature and caused by the exponential dependence between soil tempera-
9988
ture and soil respiration. The corresponding soil moisture value was determined using
the regression function between soil temperature and soil moisture. Standardization
causes the interval of soil respiration to shrink to 0.15–0.23 gCO2m
−2 h−1. The pre-
dicted standard error as well as the RMSE are smaller compared to the interpolation
based on average soil respiration alone (Fig. 6f). Comparing the maps of average and5
standardized soil respiration reveals the influence of soil temperature and soil mois-
ture. Whereas soil temperature is mainly evenly distributed, the map predominantly
shows the distribution of soil moisture deficits or excessive supply which impedes soil
respiration.
The annual flux calculated solely from 81 data points of the random series10
yielded a soil respiration rate of 2285.5 gCO2m
−2 a−1(±480.2 gCO2m−2 a−1) or
6.23MgCha−1 a−1(±1.31MgCha−1 a−1) in 2006 and can be trusted at a confidence
level of 90% within an error limit of 10% (or 99% confidence level at an error limit of
20%). Using the geostatistical approach accounting for the spatial distribution of the
plots yielded a soil respiration rate of 1875.5 gCO2m
−2 a−1 or 5.11MgCha−1 a−1 in15
2006.
4 Discussion
Considerable spatial heterogeneity of soil respiration rates was apparent within our
study site due to a variety of confounding factors:
Soil temperature is an important control for the seasonal variation of soil respiration20
(Davidson et al., 1998; Janssens and Pilegaard, 2003; Reichstein et al., 2008). Re-
ichstein et al. (2008) describe a variety of models reflecting a nonlinear positive rela-
tionship between soil respiration and soil temperature. Here, we used the Q10 function
(van’t Hoff, 1898). It is widely accepted and often used (Janssens and Pilegaard, 2003;
Hashimoto, 2005; Pavelka et al., 2007). However, it must be accepted that the Q1025
function integrates several controls: An increase in soil temperature influences many
biological and physiochemical processes e.g. root biomass distribution, litter inputs and
9989
the activity of microbial populations (Janssens and Pilegaard, 2003). However, the cal-
culated Q10-value of 3.9 fits well within the range of other published data of temperate
deciduous forests (Raich and Schlesinger, 1992; Davidson et al., 1998; Janssens and
Pilegaard, 2003; van Miegroet et al., 2007). The relationship between soil temperature
and soil respiration was much closer at dry conditions (r2=0.69, Q10=4.1) compared5
to wet conditions (r2=0.32, Q10=2.3). This is in contradiction to results by Janssens
et al. (2003). They also found that Q10 is not constant during the year, while interpret-
ing data from a Danish beech forest. However, they observed the lowest Q10 values
in early autumn during a period of drought stress when soil respiration was completely
unrelated to soil temperature and the highest values in winter. Very low or very high10
soil water contents may limit soil respiration (e.g., Davidson et al., 1998; Reichstein and
Beer, 2008; Sotta et al. 2006). In our data-set we did not experience drought. Instead,
high soil moisture values occurred at all seasons. This might be due to pockets of high
clay content (Knohl et al., 2008) with increased soil moisture. Taking this into consid-
eration, the low Q10 value in our study is more likely caused by the characteristics of15
a few collars throughout the entire year than a characteristic of seasonality. We find
no considerable relationship between soil moisture and soil respiration. That might be
caused by the relatively low number of potentially impeding situations (drought or high
soil moisture) as also found by Knohl et al. (2008) in the same study area.
4.1 Spatial heterogeneity of soil respiration20
Soil respiration as well as soil moisture did not show any spatial trends. However,
soil respiration rates and soil moisture revealed a patchy pattern without any obvious
structures. The spatial variability of soil temperature apparently does not drive this
heterogeneity because it is very small (less than ±1◦C during a single measurement
day). This is in line with Yim et al. (2003) who report that soil temperature is not25
a dominant factor for the spatial variability of soil respiration in temperate broadleaved
forests.
However, combining soil temperature and soil moisture effects within a single expo-
9990
nential equation slightly enhances the predictability of soil respiration measurements
(r2=0.63 compared to r2=0.61). But the effect was not as pronounced as described by
Knohl et al. (2008) for the same study site. Nevertheless, we used Eq. (5) to calculate
standardized soil respiration measurement values for defined averaged parameters to
achieve a higher level of accuracy.5
Due to the relatively smaller coefficient of variation of CO2 flux in the field compared
with N2O fluxes, generally less attention has been paid to the spatial heterogeneity
of CO2 fluxes (Ishizuka et al., 2005). A wide variety of authors performed statistical
analyses based on the Student’s t-distribution to calculate the number of measure-
ment locations required for an adequate level of precision (Folorunso and Rolston,10
1984; Davidson et al., 2002; Yim et al., 2003; Adachi et al., 2005; Knohl et al., 2008).
According to our calculations a higher number of plots is demanded to achieve a suf-
ficient level of precision compared to published studies within temperate deciduous
broadleaved forest. Especially the differences to the results of Knohl et al. (2008)
are remarkable, because they performed their analysis within the same research area.15
Knohl et al. used data from 10 different studies within the same research site, but
most of these measurements were performed using much less sampling points, within
smaller observational windows and only during the vegetation period. This suggests
that either there sampling density was insufficient, or – and more likely – that spatial
heterogeneity is increasing when all-season measurements are considered. This has20
important implications for the geospatial interpolation.
Soe and Buchmann (2005), also working in the Hainich National Park, observed that
trees have a stronger stabilizing effect on patterns of soil respiration than understory
vegetation within an annual perspective. If the summer or winter season were con-
sidered alone, the variation in soil respiration was much higher (ibid.). Soil respiration25
was higher in areas with high mean dbh, compared to areas with many small trees. No
correlation was found between the distance to the nearest trees and soil respiration,
but they determined a negative correlation between the number of tree species and
soil respiration.
9991
In contrast to Soe and Buchmann (2005), who worked in an old growth forest, Saiz
et al. (2006) found significantly higher soil respiration rates in close vicinity to tree
stems for the younger stands, which may be explained by the more uneven distribution
of roots in the young forests (Saiz et al., 2006). Shibistova et al. (2002) determined
significantly lower efflux rates in small forest gaps in a boreal forest, caused by recurring5
fire events which led to the total absence of higher plants. Khomik et al. (2006) showed
that spatial variability is driven by seasonal respiratory activities of roots in a 74 yr old
stand in Ontario. This could be an explanation for the limited spatial variability of efflux
rates during winter. Furthermore, broadleaved deciduous trees showed a 15% higher
respiration rate than evergreen coniferous species, whereas this difference was not10
always significant (Khomik et al., 2006).
4.2 Optimizing the parameters for interpolation
The influence of different sample sizes was assessed using nested groupings of plots
that are arranged in two of the grid cells. Huge differences between the average soil
respiration rates of different sampling densities can be caused by single outliers out-15
side the considered grid cells. This led us to vary the lag distances and bin sizes
when calculating the semivariograms (Franklin and Mills, 2003). The observed range
sizes clearly depend on the selected bin sizes. Large lag sizes (>10m) sometimes
correspond with large range sizes indicating that some drivers of soil respiration may
vary on coarser scales than others. For this study, small bin sizes were selected to20
determine autocorrelation at small ranges where large bin sizes did not indicate any
autocorrelation, and small bin sizes were selected to consider even large variability be-
tween locations close together. However, the calculated ranges of autocorrelation also
varied considerably between measurement dates. This may be caused by fluctuations
in spatial patterns due to changes in microbial activity, organic matter input or root ac-25
tivity during the course of the year (Saiz et al., 2006). The then determined average
range of the semivariograms (9m) is comparable to ranges found in investigations at
the same study site with a range of 6m (Soe and Buchmann, 2005) and in other stud-
9992
ies (e.g., 10m in Ishizuka et al., 2005). The increased amount of plots of the random
series (81) compared to the stratified random plots (43) reduces the average distances
and considerably enhances the validity of the kriging result proved by the associated
predicted standard error map (Fig. 6d). The predicted standard error decreased from
0.07 gCO2m
−2 h−1 to 0.05 gCO2m
−2 h−1 (at average respiration rates of 0.27, respec-5
tively 0.26 gCO2m
−2 h−1). This results because the average distances between the
plots of the random series better fit to the calculated range of the semivariogram than
to the stratified random series.
Single hotspot soil respiration events can adulterate not only the semivariograms but
also large areas within the interpolated maps especially when the sampling density is10
small. The results of our nested approach measurements clearly show the misleading
effect of single outliers (Fig. 4). Therefore, their elimination is important for the reliability
of the interpolation results, even when the local hot spot measurement may be func-
tionally justified. This was also supported by the estimation of an adequate number of
collars based on the Student’s t-distribution required for an adequate level of precision,15
which is widely used in previous studies (Folorunso and Rolston, 1984; Davidson et al.,
2002; Yim et al., 2003; Adachi et al., 2005; Knohl et al., 2008).
Another important factor regarding the semivariogram calculation might be the area
covered by the closed chamber used for respiration measurement. Because a larger
variation in emitted gas is covered with larger chambers these may require fewer sam-20
pling points compared to small chambers where the variation within one chamber is
much smaller (Yim et al., 2003). The chamber used in the present study has an area of
78 cm2, which is comparably small. However, with even smaller soil cores covering an
area of 1.8 cm2 Robertson et al. (1997) and Stoyan et al (2000) found autocorrelation
distances of less than 30 cm. Seen in relation to our results a possible influence of the25
chamber size becomes obvious.
9993
4.3 Interpolated maps
To avoid the influence of seasonal effects we based the kriging on mean annual soil
respiration rates. Ordinary kriging was preferred to cokriging because of the biva-
lent nature of the supporting parameters (Pringle and Lark, 2006). Soil temperature
primarily accounts for seasonal changes and did not show any spatial heterogeneity.5
Therefore, it is a poor predictor for spatial interpolation (Yim et al., 2003). Soil moisture
varies on the seasonal scale as well as on the spatial scale and the values show little
autocorrelation. Therefore, the proposed approach to standardize soil respiration using
defined soil moisture and temperature values seems to be more promising compared
to cokriging. This should not be seen as contradictory to coregionalisation attempts.10
A particular model may reproduce the components of variation better at some spatial
scales than it does at others (Pringle and Lark, 2006). Thus, cokriging methods may be
the best choice when parameters with a better autocorrelation structure are available.
But in the case of complicated autocorrelation situations like in the presented data, the
standardization approach might yield better results.15
4.4 The standardisation approach
The standardisation of soil respiration using soil temperature and soil moisture largely
reduces the interpolated spatial heterogeneity of soil respiration. The remaining het-
erogeneity must therefore be caused by other, not measured variables. As a result, the
interpolation of soil respiration data requires large sampling densities if better, spatially20
autocorrelated parameters than soil moisture and soil temperature are not available.
Standardizing of soil respiration can help to minimize artifacts caused by temporal and
spatial small scale heterogeneity of soil moisture and soil temperature. However, to
further increase the reliability of the interpolation it would be necessary to incorporate
other parameters with less temporal variability and larger autocorrelation ranges.25
In summary, the geostatistical approach yielded slightly lower results than the sta-
tistical approach, which indicates that the calculated fluxes of even a high number of
9994
measurement plots can be further improved by involving spatial parameter.
5 Conclusions
This study concentrates on the plot scale spatial effects, the relationships and con-
founding co-variations between soil temperature and soil moisture. Variations on the
micro scale as well as at larger scales have to be considered whenever plot scale varia-5
tions are influenced (Reichstein and Beer, 2008). We conclude that only a combination
of a randomized and a double nested sampling approach as well as the appropriate
incorporation of additional parameters like soil temperature and soil moisture data may
permit a sufficient estimation of the range of soil respiration, which is a prerequisite for
reliable maps of soil respiration based on interpolation. Appropriate range distances10
should always govern the sampling design. Therefore, there is a need to estimate the
range distance based on extensive pre-studies with nested measurement locations.
Furthermore, our results show that the consideration of seasonal controls facilitates
the interpolation of soil respiration measurements.
15
This Open Access Publication is funded by the University of Go¨ttingen.
References
Adachi, M., Bekku, Y. S., Konuma, A., Kadir, W. R., Okuda, T., and Koizumi, H.: Required
sample size for estimating soil respiration rates in large areas of two tropical forests and of20
two types of plantation in Malaysia, Forest Ecol. Manag., 210, 455–459, 2005.
9995
Baggs, E. M.: Partitioning the components of soil respiration: a research challenge, Plant Soil,
284, 1–5, 2006
Bouma, T. J. and Bryla, D. R.: On the assessment of root and soil respiration for soils of
different textures: interactions with soil moisture contents and soil CO2 concentrations, Plant
Soil, 227, 215–221, 2000.5
Davidson, E. A., Belk, E., and Boone, R. D.: Soil water content and temperature as independent
or confounded factors controlling soil respiration in a temperate mixed hardwood forest, Glob.
Change Biol., 4, 217–227, 1998.
Davidson, E. A., Savage, K., Verchot, L. V., and Navarro, R.: Minimizing artifacts and biases in
chamber-based measurements of soil respiration, Agr. Forest Meteorol., 113, 21–37, 2002.10
Dixon, R. K., Brown, S., Houghton, R. A., Solomon, A. M., Trexler, M. C., and Wisniewski, J.:
Carbon pools and flux of global forest ecosystems, Science, 263, 185–190, 1994.
Folorunso, O. A. and Rolston, D. E.: Spatial variability of field-measured denitrification gas
fluxes, Soil Sci. Soc. Am. J., 48, 1214–1219, 1984.
Franklin, R. B. and Mills, A. L.: Multi-scale variation in spatial heterogeneity for microbial com-15
munity structure in an eastern Virginia agricultural field, Fems Microbiol. Ecol., 44, 335–346,
2003.
Gomoryova, E.: Small-scale variation of microbial activities in a forest soil under a beech (Fagus
sylvatica L.) stand, Pol. J. Ecol., 52, 311–321, 2004.
Goovaerts, P.: Geostatistics in soil science: state-of-the-art and perspectives, Geoderma, 89,20
1–45, 1999.
Hanson, J., Edwards, N. T., Garten, C. T., and Andrews, J. A.: Separating root and soil microbial
contributions to soil respiration: a review of methods and observations, Biogeochemistry, 48,
115–146, 2000.
Hashimoto, S.: Q10 values of soil respiration in Japanese forests, J. Forest Res.-Jpn., 10, 409–25
413, 2005.
Houghton, R. A. and Woodwell, G. M.: Global climatic-change, Sci. Am., 260, 36–44, 1989.
Ishizuka, S., Iswandi, A., Nakajima, Y., Yonemura, L., Sudo, S., Tsuruta, H., and Muriyarso, D.:
Spatial patterns of greenhouse gas emission in a tropical rainforest in Indonesia, Nutr. Cycl.
Agroecosys., 71, 55–62, 2005.30
Janssens, I. A. and Pilegaard, K.: Large seasonal changes in Q10 of soil respiration in a beech
forest, Glob. Change Biol., 9, 911–918, 2003.
Kalbitz, K., Solinger, S., Park, J. H., Michalzik, B., and Matzner, E.: Controls on the dynamics
9996
of dissolved organic matter in soils: a review, Soil Sci., 165, 277–304, 2000
Khomik, M., Arain, M. A., and McCaughey, J. H.: Temporal and spatial variability of soil respi-
ration in a boreal mixedwood forest, Agr. Forest Meteorol., 140, 244–256, 2006.
Knohl, A., Soe, A. R. B., Kutsch, W. L., Gockede, M., and Buchmann, N.: Representative
estimates of soil and ecosystem respiration in an old beech forest, Plant Soil, 302, 189–202,5
2008.
Kuzyakov, Y.: Sources of CO2 efflux from soil and review of partitioning methods, Soil Biol.
Biochem., 38, 425–448, 2006
Lankreijer, H., Janssens, I. A., Buchmann, N., Longdoz, B., Epron, D., and Dore, S.: Mea-
surement of Soil Respiration, in: Fluxes of Carbon, Water and Energy of European Forests,10
edited by: Valentini, R., Ecological Studies, 163, Springer, Berlin, 37–59, 2003.
Law, B. E., Goldstein, A. H., Anthoni, M., Unsworth, M. H., Panek, J. A., Bauer, M. R., et al.:
Carbon dioxide and water vapor exchange by young and old ponderosa pine ecosystems
during a dry summer, Tree Physiol., 21, 299–308, 2001.
Martin, J. G. and Bolstad, V.: Annual soil respiration in broadleaf forests of northern Wisconsin:15
influence of moisture and site biological, chemical, and physical characteristics, Biogeo-
chemistry, 73, 149–182, 2005.
Pavelka, M., Acosta, M., Marek, M. V., Kutsch, W., and Janous, D.: Dependence of the Q10
values on the depth of the soil temperature measuring point, Plant Soil, 292, 171–179, 2007.
Pregitzer, K. S. and Euskirchen, E. S.: Carbon cycling and storage in world forests: biome20
patterns related to forest age, Glob. Change Biol., 10, 2052–2077, 2004.
Pringle, M. J. and Lark, R. M.: Spatial analysis of model error, illustrated by soil carbon dioxide
emissions, Vadose Zone J., 5, 168–183, 2006.
Raich, J. W. and Schlesinger, W. H.: The global carbon-dioxide flux in soil respiration and its
relationship to vegetation and climate, Tellus B, 44, 81–99, 1992.25
Reichstein, M. and Beer, C.: Soil respiration across scales: the importance of a model-data
integration framework for data interpretation, J. Plant Nutr. Soil Sc., 171, 344–354, 2008.
Robertson, G. P., Klingensmith, K. M., Klug, M. J., Paul, E. A., Crum, J. R., and Ellis, B. G.: Soil
resources, microbial activity, and primary production across an agricultural ecosystem, Ecol.
Appl., 7, 158–170, 1997.30
Ryan, M. G. and Law, B. E.: Interpreting, measuring, and modeling soil respiration, Biogeo-
chemistry, 73, 3–27, 2005
Saiz, G., Green, C., Butterbach-Bahl, K., Kiese, R., Avitabile, V., Farrell, E. P.: Seasonal and
9997
spatial variability of soil respiration in four Sitka spruce stands, Plant Soil, 287, 161–176
2006.
Salimon, C. I., Davidson, E. A., Victoria, R. L., and Melo, A. W. F.: CO2 flux from soil in pastures
and forests in southwestern Amazonia, Glob. Change Biol., 10, 833–843, 2004.
Schimel, D. S.: Terrestrial ecosystems and the carbon cycle, Glob. Change Biol., 1, 77–91,5
1995.
Soe, A. R. B. and Buchmann, N.: Spatial and temporal variations in soil respiration in relation
to stand structure and soil parameters in an unmanaged beech forest, Tree Physiol., 25,
1427–1436, 2005.
Sotta, E. D., Veldkamp, E., Guimaraes, B. R., Paixao, R. K., Ruivo, M. L. P., and Almeida, S. S.:10
Landscape and climatic controls on spatial and temporal variation in soil CO2 efflux in an
Eastern Amazonian Rainforest, Caxiuana, Brazil, Forest Ecol. Manag., 237, 57–64, 2006.
Stoyan, H., De-Polli, H., Bohm, S., Robertson, G. P., and Paul, E. A.: Spatial heterogeneity of
soil respiration and related properties at the plant scale, Plant Soil, 222, 203–214, 2000.
Subke, J. A., Inglima, I., and Cotrufo, M. F.: Trends and methodological impacts in soil CO215
efflux partitioning: a metaanalytical review, Glob. Change Biol., 12, 921–943, 2006
Valentini, R.: Fluxes of Carbon, Water and Energy of European Forests, Ecological Studies
163, Springer, Berlin, 2003.
van Miegroet, H., Moore, T., Tewksbury, C. E., and Nicholas, N. S.: Carbon sources and sinks in
high-elevation spruce-fir forests of the Southeastern US, Forest Ecol. Manag., 238, 249–260,20
2007.
van’t Hoff, J. H.: Lectures on Theoretical and Physical Chemistry. Part 1: Chemical Dynamics,
Ewald Arnold, London, 1898.
Vogel, J. G., Valentine, D. W., and Ruess, R. W.: Soil and root respiration in mature Alaskan
black spruce forests that vary in soil organic matter decomposition rates, Can. J. Forest Res.,25
35, 161–174, 2005.
Yim, M. H., Joo, S. J., Shutou, K., and Nakane, K.: Spatial variability of soil respiration in a larch
plantation: estimation of the number of sampling points required, Forest Ecol. Manag., 175,
585–588, 2003.
9998
Table 1. Required number of soil respiration measurement points to achieve a precision of
±10, ±20, or ±30% around full population mean at various confidence levels (80–99%).
Interval [%] 99% 95% 90% 80%
±10 288±213 140±103 92±68 52±39
±20 72±53 35±26 23±17 13±10
±30 32±24 16±11 10±8 6±4
9999
Ë
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
#
#
#
#
#
#
#
# #
#
#
#
#
#
##
#
##
#
*
#
*
#
*
#
*
#
*
#
*
#
*
#
*
#
*
#
*
#
*
#
*
#
*

























 
  

































 




 


0 25 50 75 10012.5
m
large nests
small nests
small groupings
large groupings
fetch area of eddy tower
eddy
covariance
tower
random plots
stratified random plots
¹
Fig. 1. Distribution of sampling locations in two series. Sampling locations of the random
series (81 open circles) are distributed completely random across the site. Sampling locations
of the nested series (43 dark circles and 36 triangles and 34m grid squares) are arranged in
a stratified random way in 2 nested groupings: Two of the cells were chosen randomly from the
grid. In both, 9 plots are distributed randomly (large groupings). Both cells are further divided
into quarters and in one quarter 9 plots are placed at random (small groupings).
10000
010
20
30
40
50
So
il M
oi
st
ur
e 
[%
Vo
l]
0
2
4
6
8
10
12
14
16
18
So
il T
em
pe
ra
tu
re
 [°
C
]
250 300 350 35 85 135 185 235 285 335 20 70 120 170
Day of year, 2005-2007
-0.2
0.0
0.2
0.4
0.6
0.8
1.0
So
il R
es
pi
ra
tio
n 
[g
 C
O
2 *
m
-2
*h
-1
]
2005 2006 2007
Fig. 2. Soil moisture, soil temperature, and soil respiration during the course of the study.
The graphs base on the measurements in the 81 plots of the random series. Vertical bars
represent the 95% confidence interval. The annual cycle is clearly apparent. The soil moisture
data gap during winter 2005/2006 is caused by ground frost. Furthermore, the soil respiration
measurements show considerable heterogeneity at some of the measurement dates.
10001
 SR at soil moisture 0-30 [% Vol]
-2 0 2 4 6 8 10 12 14 16 18
0.0
0.5
1.0
1.5
2.0
2.5
S
oi
l R
es
pi
ra
tio
n 
[g
 C
O
2 *
m
-2
*h
-1
Q10=4.1; r²=0.69
 CO2 = 0.0555*exp(0.1414*x)
 SR at soil moisture >30-60 [% Vol]
 all data
0 2 4 6 8 10 12 14 16 18
Tsoil [°C]
Q10 =3.7; r² =0.55
 CO2 = 0.0828*exp(0.1305*x)
 SR at soil moisture >60-90 [% Vol]
0 2 4 6 8 10 12 14 16 18 20
Q10 =2.3; r²=0.32
 CO2 = 0.0866*exp(0.0852*x)
Fig. 3. Relationship between soil temperature and soil respiration. Q10 as well as the fit and
the parameters of the models vary for different ranges of soil moisture. The Q10 for all data is
3.9 [r2=0.60; CO2=0.0711×exp(0.1365×x)].
10002
17m
(a)
(b)
(c)
(d)
(e)
(f)
so
il 
re
sp
ira
tio
n 
[g
 C
O
2 *
m
-2
 *
h-
1 ]
te
st
 c
el
l 1
te
st
 c
el
l 2
Fig. 4. Comparison of the soil respiration interpolation based on three different sampling densi-
ties within two test cells (see methods section for details). The resulting average soil respiration
values are given below the panels. Units are gCO2m
−2 h−1. Figure (a) and (b) are kriging re-
sults using only data from the stratified random distributed plots, Fig. (c) and (d) represent
results using the stratified random plots and data from the large nests, and Fig. (e) and (f) are
supplemented with data from the small nests.
10003
lag distance [m]
se
m
iv
ar
ia
nc
e
(a) (b)
(d)(c)
06/10 2005 09/30 2005
Fig. 5. Example semivariograms of soil respiration (a, b): Bin width is 10m, all possible dis-
tances included. (c, d): Bin width=1.5m, only distances up to 150m included. Auto-correlation
does only occur at very small distances between plots.
10004




























































































































 







































































 




































































































































 







































































 







(a) 2005 nested series (c) 2006 randomized series (e) 2006 randomized series
 (standardized)
(d) 2006 randomized series (PSE) (f) 2006 randomized series (PSE)(b) 2005 nested series (PSE)
High : 0.45
Low : 0.15
High : 0.07
Low : 0.02
Soil Respiration
[g CO2 *m
-2* h-1]
Fig. 6. Interpolated maps of soil respiration measurements of the nested series (a) exhibits
a large predicted standard error – PSE (b), whereas the random distribution (c) shows a much
smaller PSE (d). The most homogeneous result was achieved by standardizing the randomized
soil respiration measurements of 2006 at 11.6◦C soil temperature and correlative 35.2% soil
moisture (e) possessing little PSE (f). Units are gCO2m
−2 h−1.
10005