Page 1
LETTERS
PUBLISHED ONLINE: 31 AUGUST 2015 |
DOI: 10.1038/NGEO2516
Soil carbon storage controlled by interactions
between geochemistry and climate
Sebastian Doetterl
1,2
*
, Antoine Stevens
3,4
, Johan Six
4
, Roel Merckx
5
, Kristof Van Oost
3
,
Manuel Casanova Pinto
6
, Angélica Casanova-Katny
7
, Cristina Muñoz
8
, Mathieu Boudin
9
,
Erick Zagal Venegas
8
and Pascal Boeckx
1
Soils are an important site of carbon storage
1
. Climate is
generally regarded as one of the primary controls over soil
organic carbon
1,2
, but there is still uncertainty about the
direction and magnitude of carbon responses to climate
change. Here we show that geochemistry, too, is an important
controlling factor for soil carbon storage. We measured a range
of soil and climate variables at 24 sites along a 4,000-km-long
north–south transect of natural grassland and shrubland in
Chile and the Antarctic Peninsula, which spans a broad range
of climatic and geochemical conditions. We find that soils
with high carbon content are characterized by substantial
adsorption of carbon compounds onto mineral soil and low
rates of respiration per unit of soil carbon; and vice versa for
soils with low carbon content. Precipitation and temperature
were only secondary predictors for carbon storage, respiration,
residence time and stabilization mechanisms. Correlations
between climatic variables and carbon variables decreased
significantly after removing relationships with geochemical
predictors. We conclude that the interactions of climatic and
geochemical factors control soil organic carbon storage and
turnover, and must be considered for robust prediction of
current and future soil carbon storage.
Soil organic carbon (SOC) is one of the most important terrestrial
C pools, with spatially variable but large annual C exchanges with
the atmosphere
1,2
. Single and interactive effects of climatic and
biotic factors on SOC dynamics have been studied intensively at
various spatial and temporal scales, but are still poorly represented
in current Earth System Models (ESMs; refs 3–6). Generally, climatic
factors have been regarded as primary controls in empirical and
modelling approaches
1,2,7
. Consequently, ESMs predict a significant
contribution of SOC to future climate change. However, recent
model-based observations indicate that geochemical factors very
likely play crucial roles in SOC turnover and large uncertainties
remain
8,9
. These uncertainties are explained partly by ESMs poorly
representing the current (observed) global SOC distribution
1
and
partly by inadequate parameterization of the temperature sensitivity
of SOC, microbial carbon use efficiency, and mineral surface
sorption of organic matter
1,8,9
. The latter indicates shortcomings in
the current approaches that focus on climatic controls and neglect
other factors, such as the geochemistry of soils and effects of soil
mineralogy on carbon stabilization
10,11
. The geochemistry of the
reactive mineral phase of a soil depends on the composition of
its parent material and weathering status. Weathering is crucially
driven by the time since the onset of weathering, resilience
of minerals to weathering, vegetation cover, local climate, and
hydrologic conditions. Owing to the difference in scales between
climate research and geochemical-related soil research (mostly large
and local scale, respectively), the interactions of these key factors for
SOC dynamics have rarely been assessed.
Here, we contribute to the debate on the importance of various
environmental factors for global SOC dynamics by assessing
the degree of direct versus indirect effects of geochemical and
climatic factors on SOC stocks (SOC
Stock
), concentrations (SOC
%
),
specific potential respiration rates (SPR) and SOC fractions along
an approximately 4,000 km north–south transect across Chile
and the Antarctic Peninsula spanning a large range of climatic
and geochemical parameters. All sites were under grassland
and/or shrubland vegetation
12
to keep C input as similar as
possible (Supplementary Table 2). However, differences in net
primary productivity (NPP) could not be excluded; biomass
production along the transect ranged between 800 to 4,500 kg
dry weight ha
−
1
yr
−
1
, with extreme values of 300 kg ha
−
1
yr
−
1
in hot arid climates and up to 8,000 kg ha
−
1
yr
−
1
in temperate
humid climates
13,14
.
Our data reveal strong connections between geochemical and
climatic drivers of global SOC dynamics, demonstrating a need to
consider their interactions when addressing current spatial patterns
of SOC storage and turnover, as well as future global responses
of SOC to climate change. We sampled topsoil (0–10 cm) at 24
sites, in triplicate, along this transect (Supplementary Table 1). To
study the effects of geochemistry on the selected SOC variables
(SOC
%
, SOC
Stock
, SPR and SOC fractions), we determined the
following key geochemical characteristics of the soil samples: base
saturation (BS) of the potential cation exchange capacity (CEC
pot
);
soil texture; contents of silicon (Si) and abundant metals (Fe, Mn,
Al); Si/Al ratio; the total reserve in ‘base’ cations (TRB); and
geochemical soil fertility parameters (pH
KCl
, P, K) (Supplementary
Table 3). We used mean annual precipitation and temperature as
indicators of climatic conditions (see Supplementary Fig. 3). The
predicting variables used here are thus proxies for environmental
conditions controlling SOC dynamics (for a detailed discussion
see Supplementary Information). We assume that geochemical
1
Isotope Bioscience Laboratory, Ghent University, 9000 Gent, Belgium.
2
Department of Geography, University of Augsburg, 86159 Augsburg, Germany.
3
George Lemaître Centre for Earth and Climate Research, Earth & Life Institute, Université catholique de Louvain, 1348 Louvain-la-Neuve, Belgium.
4
Department of Environmental Systems Science, ETH Zurich, 8092 Zurich, Switzerland.
5
Department of Earth and Environmental Sciences, Katholieke
Universiteit Leuven, 3001 Heverlee, Belgium.
6
Departamento de Ingeniería y Suelos, Universidad de Chile, 8820808 Santiago, Chile.
7
Facultad de
Química y Farmacia, Universidad de Santiago de Chile, 8820808 Santiago, Chile.
8
Departamento de Ciencias del Suelo y Recursos Naturales, Universidad
de Concepción, 3812120 Chillán, Chile.
9
Royal Institute for Cultural Heritage, 1000 Bruxelles, Belgium.
*
e-mail:
Sebastian.Doetterl@Ugent.be
780
NATURE GEOSCIENCE
| VOL 8 | OCTOBER 2015 |
www.nature.com/naturegeoscience
© 2015 Macmillan Publishers Limited. All rights reserved
Page 2
NATURE GEOSCIENCE
DOI: 10.1038/NGEO2516
LETTERS
Predominant climate
0
5
10
0
10
20
30
SOC
Stock
kg m
−2
(0−10 cm)
Respiration (SPR)
µ
g CO
2
−C g
−1
SOC h
−1
Mediterranean
-(semi-)arid
Temperate
semi-oceanic
Subpolar
(semi-)oceanic
Subpolar arid
Temperate oceanic
Drake Passage
Drake Passage
Antarctic
maritime
WRB soil orders
DU AR
AN
AN
AN
CM
GL
CM
KS
CR
LP
AC
CM
CH/KS/PH
CM/AN/KS/LP
Latitude
30
°
S
35
°
S
40
°
S
45
°
S
50
°
S
55
°
S
60
°
S
Geoclimatic condition
Harsh conditions
Mechanical weathering
dominant
Mild conditions
Chemical weathering
dominant
POM accumulation
Harsh conditions
Mechanical weathering
dominant
POM accumulation
Effect on SOC
Decreasing SOC stock
Increasing specific respiration
Decreasing SOC stock
Increasing specific respiration
Organo-mineral
association
SOC
Microbial
SOC
Mineral
1%
89%
98%
43%
Relative mean standard deviation
SOC
Stock
±
13%
SOC
Mineral
±
15% SOC
Microbial
±
70%
Respiration
(SPR)
±
17%
Figure 1
| SOC response variables in geographical context.
Relationship between soil organic carbon stocks (SOC
Stock
, left-hand
y
axis), specific potential
respiration (SPR, right-hand
y
axis), mineral-associated SOC (SOC
Mineral
, green circles), microbially available SOC (SOC
Microbial
, blue circles) and latitude
(
x
axis). The position of the points relates to the value on the relevant
y
axis, and the size of the accompanying circle relates to SOC
Microbial
and SOC
Mineral
as a percentage of SOC
Stock
. Relative uncertainty is calculated as the mean relative standard deviation of each SOC response variable. An interactive map
showing further information on the SOC response and prediction variables in their geographic context is available online (see Supplementary Data). WRB
soil orders: DU, Durisol; AR, Arenosol; CH, Chernozem; KS, Kastanozem; PH, Phaeozem; AN, Andisol; CM, Cambisol; GL, Gleysol; AC, Acrisol; LP, Leptosol;
CR, Cryosol.
Table 1 |
Relative contributions of geoclimatic variables in predicting SOC response variables.
MAP MAT Clay Silt
Si
BS
Al
TRB
pH
KCl
Fe
K
P
Mn
R
2
cv
RMSE
cv
RMSE unit
Least angle linear regression (LARS)
SOC
%
5.07 NS
NS
NS
−
127.78
−
10.75
−
72.35
−
36.97
NS
−
25.62
−
13.08 NS NS 0.95 11.54
SOC g kg
−
1
SOC
Stock
1.87 NS
NS
NS
−
1.61
NS
NS
−
2.69
NS
NS
−
1.18
NS NS 0.75 1.3
SOC kg m
2
SPR
−
2.75 NS
NS
NS
5.38
4.58
NS
NS
NS
NS
1.51
NS NS 0.76 5
µ
g CO
2
–C g
−
1
SOC h
−
1
Best subset selection linear regression (BSS)
SOC
%
NS
NS
NS
NS
−
0.93
NS
−
0.47
NS
NS
−
0.36
NS
NS NS 0.96 11.24
SOC g kg
−
1
SOC
Stock
NS
NS
NS
NS
−
0.56
NS
NS
−
0.60
NS
NS
NS
NS NS 0.71 1.22
SOC kg m
2
SPR
NS
NS
NS
NS
0.52
0.52
NS
NS
NS
NS
NS
NS NS 0.86 3.13
µ
g CO
2
–C g
−
1
SOC h
−
1
Standardized coefficients indicating variable importance in the LARS and BSS prediction models (see Methods).
R
2
and RMSE are computed through cross-validation. MAP, mean annual precipitation;
MAT, mean annual temperature; clay, silt, soil clay and silt content; TRB, total reserve in ‘base’ cations; BS, base saturation of potential cation exchange capacity; Si, Al, Fe, K, P, Mn, soil total content of
tested elements; pH
KCl
, soil pH measured in KCl solution; blanks, not selected; SPR, specific potential respiration; SOC
%
, SOCconcentration; SOC
Stock
, SOC stock (0–10 cm depth); NS, not significant.
differences among soils in the same climate zone are mainly
related to the geochemical nature of the parent material and its
weathering status
15
. To eliminate the effect of climate on specific
potential respiration, we conducted incubation experiments under
controlled conditions where temperature and moisture conditions
were similar for all samples and only geochemistry varied. To assess
the contributions of climatic and geochemical factors to the control
of SOC response variables, we used multivariate modelling and
(partial) correlations (see Methods).
The respiration data show that soils with high SOC
Stock
generally
respired CO
2
at lower rates per unit C during a 50-day incubation
than soils with low SOC contents (Fig. 1 and Supplementary
Data Table 3). The quartile of samples with the lowest SOC
Stock
(1.4
±
0.3 kg C m
−
2
(0–10 cm depth)), predominantly from arid
regions with extremely high or low annual temperatures, respired
2.4 times more than the quartile with the highest SOC
Stock
(6.8
±
0.8 kg C m
−
2
(0–10 cm depth)), predominantly from humid
regions with mild to warm temperatures. In addition, about
38
±
15% of SOC
Stock
in the lowest quartile soils is microbially
available (SOC
Microbial
), whereas this is only 13
±
15% for the
highest quartile. Furthermore, in humid regions with mild to warm
temperatures, where (bio)chemical weathering rates are high, SOC
is predominantly found in association with the mineral fraction
(SOC
Mineral
; see Methods and Supplementary Information 1.4 for
details on SOC fractionation results and interpretation). Hence,
organo-mineral interactions create physico-chemical barriers for
microorganisms to access SOC sources, leading to large, protected
SOC stocks in the humid regions with mild to warm temperatures.
In contrast, in arid regions with extremely high or low temperatures,
where soil weathering is largely driven by physical factors
(temperature-driven mechanical shredding of minerals)
16,17
, less
SOC is associated with minerals. This finding, together with the
high SPR and SOC
Microbial
, suggests that physico-chemical protection
of SOC from decomposers is limited in arid soils. Our results
show that geochemical parameters have the highest prediction
power for the investigated SOC response variables (Table 1). The Si
NATURE GEOSCIENCE
| VOL 8 | OCTOBER 2015 |
www.nature.com/naturegeoscience
781
© 2015 Macmillan Publishers Limited. All rights reserved
Page 3
LETTERS
NATURE GEOSCIENCE
DOI: 10.1038/NGEO2516
0.21
0.21 −0.04 −0.28
0.2
0.34
0.07
0.21
0.48
∗
0.04
0.4
0.21
−0.01
−0.1
−0.1
0.17
0.17
−0.08
−0.39 0.06
−0.11
−0.31 0.06
−0.2
−0.11
0.17
−0.15
−0.14
0.11
0.18
−0.13
−0.42 −0.04
−0.13 −0.31
0.01 −0.26 −0.18
−0.05
−0.77
∗
−0.79
∗
−0.6
∗
−0.71
∗
−0.81
∗
−0.73
∗
−0.75
∗
−0.79
∗
−0.7
∗
−0.59
∗
−0.42 −0.79
∗
−0.07
0.72
∗
0.71
∗
0.53
∗
0.66
∗
0.72
∗
0.66
∗
0.68
∗
0.75
∗
0.64
∗
0.54
∗
0.45
0.7
∗
0.03
0.74
∗
0.74
∗
0.57
∗
0.68
∗
0.74
∗
0.69
∗
0.75
∗
0.74
∗
0.69
∗
0.57
∗
0.53
∗
0.74
∗
−0.1
Temperature (MAT)
Precipitation (MAP)
Respiration
(SPR)
SOC
Stock
SOC
%
Respiration
(SPR)
SOC
Stock
SOC
%
Zer
o-or
der
Cla
y
Silt
Si
Al
BS
TRB
pH_K
Cl
Fe
K
Mn
P
Geochem
Partial correlation controls
(Partial)
correlation
Pearson’s
r
0.8
0.4
0.0
−0.4
−0.8
Figure 2
| Partial correlations between SOC response and climatic variables.
Change of correlation between SOC response and climatic variables
controlled for geochemical variables separately and all combined (column ‘Geochem’). Difference between zero-order and partial correlations indicate the
level of dependency of a given predictor and the SOC response. The colour and numbers shown indicate the strength and sign of the correlation. (No
change in colour between controlled variable and zero-order
=
no dependency; decrease/increase of colour intensity
=
loss of/gain of correlation.)
Significance of the correlations (
∗
) is evaluated at the 0.05 level. Abbreviations are explained in the caption to Table 1.
concentration was a key predictor and the only variable maintained
in all models. In contrast, precipitation was generally of minor
importance in all models combining geochemical and climatic
predictors; temperature was not important in any model.
Further support for the importance of geochemical soil proper-
ties is derived from the partial correlation analysis (see Methods for
details). Although the correlations between precipitation and SOC
response variables are highly significant (zero-order correlations
in Fig. 2) the partial correlation analysis shows that the Pearson
correlation coefficients between precipitation and SOC
%
, SOC
Stock
and SPR response variables decreased significantly after removing
correlations with all geochemical predictors. In sharp contrast, no
such strong decrease was observed in the strength of the correlations
between SOC response variables and most geochemical predictors
(except for BS and Si) when removing correlations with all climatic
predictors (Supplementary Data Table 8). Annual temperature had
no effect on the prediction power of geochemical variables (see
Fig. 2 and Supplementary Fig. 2). However, annual temperature
played a slightly larger role when controlling for the effect of all
geochemical predictors combined, as the correlation of temperature
with SOC variables increased slightly when controlling for the effect
of geochemical variables (Fig. 2). All of these results indicate that
SOC response variables partly depend on climatic variables, inde-
pendently of geochemical soil conditions, but that these relation-
ships are much weaker than those between geochemical predictors
and SOC response variables. This conclusion is supported by the
identified importance of Si, for which high values are indicative of
high SPR and low SOC
Stock
, and related to either a lower degree of
chemical weathering or felsic geochemistry of the parent material.
Overall, precipitation showed high correlations with SOC response
variables, but was only a weak predictor of SOC response variables
owing to the dominance of geochemical variables. Hence, our re-
sults clearly indicate that, in the long-term, climatic variables have
predominantly indirect effects on SOC dynamics via their influence
on soil geochemistry.
Undoubtedly, climatic variables directly control the SOC cycle at
shorter annual timescales, through processes such as heterotrophic
CO
2
respiration and photosynthetic CO
2
fixation. Hence, current
research on global scales focuses on changes in NPP and C inputs to
soils to assess the consequences of global change. However, we argue
that climate is not the main control for SOC storage on decadal to
centennial timescales (see Supplementary Information 1.4) during
which the SOC storage potential evolves because of mineral
weathering. Instead climate acts largely as a driving factor through
governing soil weathering, which results in geochemical changes in
soils that directly control SOC stabilization in the longer term. In
topsoils, such as the ones investigated here, high SOC stocks are
driven predominantly through physico-chemical stabilization of C
with minerals (see Supplementary Information 1.5). Climate change
on decadal timescales will probably affect NPP and decomposition,
but the effects of changing inputs/outputs of C to soils are
constrained by geochemical soil properties.
Hence, our data clearly shows that soil geochemistry is a
crucial factor for predicting SOC stocks and dynamics at larger
scales. Thus, uncertainty around the global distribution and climate
feedback of SOC is probably due to inadequate representation of
the geochemistry of soils in current global assessments and ESMs.
Understanding the effects of climate change on SOC dynamics
also requires understanding the relationship between climate and
soil geochemistry, because the mineral matrix ultimately controls
the fate of SOC. Only recently, first steps were taken to represent
some key geochemical mechanisms controlling SOC, such as the
stabilization of C by association with minerals and/or within soil
aggregates, into ESMs applied at global scales
3,18
. In ESMs, soils are
typically considered as simple textural matrices in which heat and
water fluxes occur. However, the mechanistic interaction of SOC
with minerals is sufficiently understood
19,20
to incorporate them into
ESMs and thus further enhance their predictive power. However, in
addition to geochemical controls, this will also require an adequate
representation of SOC temperature sensitivity, the molecular
structure and recalcitrance of molecules to decomposition, potential
sources for SOC priming or nutrient limitation in soils, microbial
community structures or environmental controls (soil moisture
and oxygen limitation; see further discussion Supplementary
Information 1.2). Thus, further efforts should focus on obtaining the
required global scale biogeochemical and environmental soil data to
represent SOC stabilization mechanisms and direct process-related
controls in ESMs.
In conclusion, we present evidence that the long-term SOC
stabilization potential of soils under similar vegetation depends
782
NATURE GEOSCIENCE
| VOL 8 | OCTOBER 2015 |
www.nature.com/naturegeoscience
© 2015 Macmillan Publishers Limited. All rights reserved
Page 4
NATURE GEOSCIENCE
DOI: 10.1038/NGEO2516
LETTERS
strongly on the reactivity of the soil mineral phase. Whether
temperature or precipitation governs the response of mineral
reactivity depends on the prevailing climatic conditions under
which soils have developed and whether temperature or moisture
is limiting soil weathering (see also discussion in Supplementary
Information 1.4). Hence, analysing the geochemistry of soils formed
under natural conditions, at a global scale, could substantially
enhance our understanding of the dynamics of the global terrestrial
SOC cycle. Furthermore, including mineral-driven stabilization in
ESMs will improve our predictions of SOC feedbacks to warming
1,21
.
Data at present informing ESMs on C turnover responses to global
change are mostly derived from controlled experiments under
elevated [CO
2
] and/or with a focus on climatic and biotic factors
1,3
,
implying that minerals have limited importance for SOC dynamics.
This contradicts our findings regarding SOC variables of natural soil
systems, which reflect long-term (decades to millennia) interactions
of factors, varying in both strength and duration. If ESMs consider
climate only as a direct control of SOC dynamics, large inaccuracies
are likely to arise because predictions for a changing climate will not
account for changes in soil mineral phases.
Methods
Methods and any associated references are available in the
online
version of the paper
.
Received 15 February 2015; accepted 22 July 2015;
published online 31 August 2015
References
1. Carvalhais, N.
et al.
Global covariation of carbon turnover times with climate
in terrestrial ecosystems.
Nature
514,
213–217 (2014).
2. Shao, P., Zeng, Z., Moore, D. J. P. & Zeng, X. Soil microbial respiration from
observations and earth system models.
Environ. Res. Lett.
8,
034034 (2013).
3. Davidson, E. A. & Janssens, I. A. Temperature sensitivity of soil carbon
decomposition and feedbacks to climate change.
Nature
440,
165–173 (2006).
4. Nie, M.
et al.
Positive climate feedbacks of soil microbial communities in a
semi-arid grassland.
Ecol. Lett.
16,
234–241 (2013).
5. Karhu, K.
et al.
Temperature sensitivity of soil respiration rates enhanced by
microbial community response.
Nature
513,
81–84 (2014).
6. Feng, X., Simpson, A. J., Wilson, K. P., Williams, D. D. & Simpson, M. J.
Increased cuticular carbon sequestration and lignin oxidation in response to
soil warming.
Nature Geosci.
1,
836–839 (2008).
7. Chen, S., Huang, Y., Zou, J. & Shi, Y. Mean residence time of global topsoil
organic carbon depends on temperature, precipitation and soil nitrogen.
Glob. Planet. Change
100,
99–108 (2013).
8. Sulman, B. N., Phillips, R. P., Oishi, A. C., Shevliakova, E. & Pacala, S. W.
Microbe-driven turnover offsets mineral-mediated storage of soil carbon under
elevated CO
2
.
Nature Clim. Change
4,
1099–1102 (2014).
9. Tang, J. & Riley, W. J. Weaker soil carbon-climate feedbacks resulting from
microbial and abiotic interactions.
Nature Clim. Change
5,
56–60 (2015).
10. Torn, M. S., Trumbore, S. E., Chadwick, O. A., Vitousek, P. M. &
Hendricks, D. M. Mineral control of soil organic carbon storage and turnover.
Nature
389,
170–173 (1997).
11. Trumbore, S. E. & Torn, M. S. in
Soils and Global Change
(ed. Holland, E. A.)
1–34 (NATO Advanced Study Institute, 2003).
12. Norton, K. P., Molnar, P. & Schlunegger, F. The role of climate-driven chemical
weathering on soil production.
Geomorphology
204,
510–517 (2014).
13. Ruiz, I.
Praderas Para Chile
[in Spanish] (Ministerio de Agricultura, 1996).
14. Barcikowski, A., Lyżwtníska, R. & Zarzycki, K. Growth rate and biomass
production of
Deschampsia antarctica
Desv. in the Admiralty Bay region, South
Shetland Islands, Antarctica.
Pol. Polar Res.
20,
301–311 (1999).
15. Dosseto, A., Buss, H. L. & Suresh, P. O. Rapid regolith formation over volcanic
bedrock and implications for landscape evolution.
Earth Planet. Sci. Lett.
337–338,
47–55 (2012).
16. Goudie, A. S. & Viles, H. in
Quaternary and Recent Processes and Forms
Vol. 4
(eds Burt, T. P., Chorley, R. J., Brunsden, D., Cox, N. J. & Goudie, A. S.)
129–164 (Geological Society of London, 2008).
17. Gislason, S. R.
et al.
Direct evidence of the feedback between climate and
weathering.
Earth Planet. Sci. Lett.
277,
213–222 (2009).
18. Riley, W. J.
et al.
Long residence times of rapidly decomposable soil organic
matter: Application of a multi-phase, multi-component, and vertically-resolved
model (BAMS1) to soil carbon dynamics.
Geosci. Model Dev.
7,
1335–1355 (2014).
19. Six, J., Conant, R. T., Paul, E. A. & Paustian, K. Stabilization mechanisms of soil
organic matter: Implications for C-saturation of soils.
Plant Soil
241,
155–176 (2002).
20. von Luetzow, M.
et al.
Stabilization of organic matter in temperate soils,
mechanisms and their relevance under different soil conditions—a review.
Eur.
J. Soil Sci.
57,
426–445 (2006).
21. Zhou, T., Phi, P., Hui, D. & Luo, Y. Global pattern of temperature sensitivity of
soil heterotrophic respiration and its implications for carbon-climate feedback.
J. Geophys. Res.
114,
G02016 (2009).
Acknowledgements
This research was financed within the framework of the BELSPO IUAP project
No. P7-P24 ‘SOGLO- Soils under Global change’ (Belgium). Further funding was
provided in Chile by FONDECYT 1121138 and 1120895 and Instituto Antártico Chileno,
INACH FR0112 for sampling campaigns in Chile and the Antarctic Peninsula. K.V.O. is a
Research Associate of the Fonds de la Recherche Scientifique (FNRS), Belgium.
Author contributions
S.D., P.B. and E.Z.V. designed the research. E.Z.V., M.C.P., A.C.-K. and C.M. conducted
sampling campaigns. S.D., A.S. and M.B. collected and analysed the data. S.D., P.B. and
J.S. interpreted the data. All authors contributed to the writing of the paper.
Additional information
Supplementary information is available in the
online version of the paper
. Reprints and
permissions information is available online at
www.nature.com/reprints
.
Correspondence and requests for materials should be addressed to S.D.
Competing financial interests
The authors declare no competing financial interests.
NATURE GEOSCIENCE
| VOL 8 | OCTOBER 2015 |
www.nature.com/naturegeoscience
783
© 2015 Macmillan Publishers Limited. All rights reserved
Page 5
LETTERS
NATURE GEOSCIENCE
DOI: 10.1038/NGEO2516
Methods
Research sites, selection criteria and sampling.
Between December 2012 and
April 2013, 24 sites were sampled across Chile and the Antarctic Peninsula,
covering a vast range of geoclimatic conditions (Supplementary Table 1).
We covered all major global climate zones under which natural grassland and
shrubland evolve, including hot, temperate and cold, as well as both humid and
arid zones to provide a global inference space for our results. Sampling sites were
selected that provided: carbonate-free soil samples (null HCl reaction); varying soil
moisture regimes (aridic, ustic, xeric, udic and perudic); varying soil temperature
regimes (thermic, isothermic, mesic, isomesic and cryic); variation in key
geochemical properties of soils related to soil quality, weathering and SOC
stabilization via organo-mineral interaction (CEC, BS, total content of Si, Fe, Mn,
Al); similar vegetation; and no human disturbance (natural prairie, grassland or
shrubland)
22
. Our sites did not include wetlands, where at least part of the year
anaerobic soil conditions prevail, reducing rates of decomposition, and the
accumulation of large C stocks. Likewise, forest sites were excluded, meaning that
the factors important for accumulation of C in forest soil organic layers are not
considered. These processes are beyond the scope of our study focusing on
comparable grassland/shrubland systems. The sampled soils represent pedons of
soil series from the official regional soil surveys conducted by the Chilean Natural
Resources Information Centre (CIREN) and cover 11 WRB Soil Taxonomy groups
(Durisol; Arenosol; Chernozem; Kastanozem; Phaeozem; Andisol; Cambisol;
Gleysol; Acrisol; Leptosol; Cryosol) (Supplementary Table 1). Following published
sampling methodology
23
, 20 randomized soil cores (0 to 10 cm depth) were
collected from a randomly chosen 25
×
25 m soil plot and mixed. At each study site
bulk density was determined using Kopecky cylinders and each composed sample
(2 to 3 kg) was sieved to remove soil fauna, rocks fragments (
>
2 mm) and fine
roots, then transported in coolers to laboratories at Concepción University, where
they were stored in cooling chambers at
−
20
◦
C until analysis.
Climate and vegetation.
Mean annual temperature and precipitation data for the
sampled sites in Chile were taken from the WORLDCLIM data set
24
records for the
period 1950–2000 and for the Antarctic Peninsula from the climate records of the
nearest research stations: ‘Esperanza’ and ‘Bellingshausen’. Most grasslands in Chile
are derived from the removal of forest since the arrival of European settlers through
clearing and fire, or more recently following exotic
Pinus
or
Eucalyptus
plantations.
Chile has only a small number of native grass species and most were imported by
the Europeans. The selected study sites are not intensively managed and are used
only for extensive grazing and rangeland. Thus, establishing plant communities
similar to natural grasslands in Europe and Western Asia
25
. The principal
vegetation formations of the study transect and their main characteristics are
described in ref. 26, and dominant and secondary vegetal species were identified
for each site (Supplementary Table 2). In the north, the sparse desert coastal scrub
vegetation extends into the interior between 24
◦
and 32
◦
S, generating a transition
zone from the Atacama Desert in the north towards the winter-wet Mediterranean
climatic zone of Central Chile. It encompasses open shrub (matorral) vegetation
that gradually changes from 30
◦
S onwards to xerophytic scrub further south.
Entering into the Mediterranean climate zone, the vegetation changes to a
sclerophyllous scrub. Further south, climatic conditions with areas of high rainfall
and modest temperate regimes have led to the development of large areas with
deciduous and evergreen broadleaved forests that dominate the natural vegetation
of Central and Southern Chile. South of 47
◦
S, precipitation exceeds 4,000 mm yr
−
1
and the vegetation becomes increasingly moorland. Further south, along the low
Andes in Southern Patagonia and Tierra del Fuego, the vegetation is dominated by
gramineous steppe, with characteristics governed by a marked precipitation
gradient ranging from 4,000 mm yr
−
1
at the western side to 300 mm yr
−
1
at the
eastern side. The Antarctic Peninsula then closes the gradient of climatic variability
within the transect with harsh climatic conditions characterized by short vegetation
periods and generally low mean annual temperatures. Vegetation is generally
sparse and only two flowering plants occur along the northern and western coasts
of the Antarctic Peninsula in areas with the mildest maritime climate in the region.
Soil analysis.
pH, CEC, BS and soil texture. Soil pH was determined
potentiometrically in 25 ml 0.01 M CaCl
2
(1:2.5 soil:solution ratio) with a glass
electrode using a portable multi-parameter meter HI9828 (Hanna Instruments
US). Potential cation exchange capacity (CEC
pot
) was determined by quantifying
NH
+
4
exchanged with 2 M KCl after saturating cation exchange sites with
ammonium acetate buffered at pH 7.0 (ref. 27). Exchangeable Al was extracted by
1 M KCl solution and determined colorimetrically. The total percentage base
saturation (BS), defined as the relative availability of each cation for CEC
pot
, was
calculated as a percentage of CEC
pot
. Soil texture was measured on composite
samples of each study site using the hydrometer method following ref. 28. For
samples containing
>
5% organic C, samples were pre-treated with 10% H
2
O
2
.
TRB and total elemental composition.
We measured the total reserve in ‘base’
cations (TRB, the sum of total Ca, Na, K and Mg, in cmol
c
kg
−
1
) following
published protocols
29
. The TRB can be used to assess the degree of weathering and
the weathering potential of soils by assessing the abundance of Ca, Na, K and Mg
cations. To express the nature and amount of weatherable minerals and the amount
of metals potentially available to form organo-mineral associations, the total
elemental content (TEC) was determined for each of these cations as well as for Si,
Fe, Al and Mn, by inductively coupled plasma-optical emission spectrometry
(ICP-OES) after borate fusion. High TRB indicates a low degree of soil weathering
and low TRB a high degree of soil weathering.
Soil incubations.
We incubated three replicate samples of 80 g soil from each study
site. After a 10-day pre-incubation period, respiration was measured during a
50-day period while keeping moisture (at 60% soil water holding capacity) and
temperature (20
◦
C) constant. Each sample was put in a 1,000 ml sealed Mason jar
with no further additives, except that evaporated water was replaced during the
experiment. Thus, we established stable conditions for microbial activity to induce
respiration close to the potential maximum heterotrophic soil respiration rate
30
. To
avoid CO
2
saturation effects influencing microbial decomposition processes, the
incubation jars were flushed with fresh air after each measurement. Samples of the
gas mixture within the incubation jars were taken periodically every three to seven
days throughout the experiment and analysed for CO
2
concentration using a gas
chromatograph (Shimadzu GC-14B, Shimadzu Scientific Instruments). We
calculated specific potential respiration as CO
2
–C per unit SOC.
Soil C fractionation, C measurements and recovery rates.
A sample of 80 g of
each soil was fractionated in triplicate to derive functional SOC pools. We used a
method based on the conceptual SOC fraction model proposed by ref. 31
(Supplementary Fig. 1). Briefly, the scheme consists of a series of physical
fractionation techniques applied to isolate SOC fractions, differentiated by
stabilization mechanisms (chemical, biochemical and physical), which can also be
associated with different turnover times and SOC stability. SOC was fractionated
into coarse particulate organic matter (
>
250
µ
m), micro-aggregate-associated SOC
(250–53
µ
m), and non-aggregated silt and clay aggregated SOC (
<
53
µ
m) using a
micro-aggregate isolator.
The microbial available bulk SOC pool (SOC
Microbial
) was estimated using
oxidation with 6% sodium hypochloride according to ref. 32 and inversely related
to the mineralizable SOC fraction. Note that the fraction of SOC that is extractable
with this method cannot be purely compared to biological mineralization, as the
fraction of chemically oxidizable SOC is sensitive to soil mineralogical conditions
33
and also related to higher uncertainties between replicates (Fig. 1). Thus, we used
the method only as an additional proxy in combination with the measured specific
potential respiration (SPR) rates and the measured amount of mineral-associated
and particulate organic matter, as the investigated gradient spans a wide range of
mineral versus POM-associated C ratios.
SOC
%
was measured in 1 g ground subsamples using a dry combustion analyser
(Variomax CN, Elementar GmbH) with a measuring range of 0.2–400 mg C g
−
1
soil
(absolute C in sample) and a reproducibility of
<
0.5% (relative deviation).
Recovery rates exceeding 97% and 91% were obtained for the soil mass and C mass,
respectively, across all fractions. The isolated fractions were analysed for total SOC
using an elemental analyser (ANCA-GSL PDZ) coupled to an Isotope Ratio Mass
Spectrometer (2020, SerCon).
Incentives for applied scheme.
The gross of the annual, short-term
soil–atmosphere C fluxes is related to decomposition of labile, non-protected SOC.
SOC is not a homogeneous pool but consists of fractions with different turnover
times ranging from days to millennia. Soils can vary greatly in their potential to
sequester C owing to their physical and geochemical properties. SOC can be
stabilized against decomposition by three key mechanisms: inherent biochemical
recalcitrance, organo-mineral associations by interaction with mineral surfaces,
and physical protection by spatial separation of C from decomposers or
inaccessibility of SOC within soil aggregates (for an overview see ref. 20). Derived
from fractionating the sampled soils into functional carbon pools (Supplementary
Fig. 1), we evaluated with the acquired data the relative importance of different
geochemical parameters, taking into account the amount of microbially available C
(SOC
Microbial
) and C stabilized through interaction with minerals (SOC
Mineral
), either
within aggregates (m) or associated with free silt and clay particles (s
+
c) and as
particulate, non-mineral-associated organic C (CPOM).
Statistical analysis.
All statistical analyses were performed with R statistical
software (R 3.1.1; ref. 34) and the packages ‘leaps’, ‘lars’, ‘caret’, ‘psych’ and ‘relaimpo’.
Predicting SOC response variables and relative importance of predictors.
To
identify variables that have the greatest prediction power for the selected SOC
response variables, we used three linear and nonlinear multivariate models: best
subset (BSS), least angle regression (LARS) and piecewise linear regression
(CUBIST). BSS regression tests all possible combinations of the input variables and
finds a subset of size
k
that gives the highest prediction power at the lowest
NATURE GEOSCIENCE
|
www.nature.com/naturegeoscience
© 2015 Macmillan Publishers Limited. All rights reserved
Page 6
NATURE GEOSCIENCE
DOI: 10.1038/NGEO2516
LETTERS
prediction error (for example,
R
2
, mean squared error, or Akaike Information
Criterion). The LARS algorithm provides an efficient way to solve the least absolute
shrinkage and selection operator (LASSO) path of solutions. LASSO is a linear
regression method that minimizes the sum of squared errors with a penalty on the
sum of the absolute values of the coefficients (
`
1-norm). Effectively, setting the
penalty sufficiently high will cause some of the coefficients to shrink towards zero
and ultimately remove the corresponding variables from the set of predictors. An
overview of BSS and LARS is given in ref. 35. CUBIST is a model tree approach
combined with a nearest-neighbour correction
34
, which creates a tree structure
defined by a set of rules with linear models at the terminal nodes. Cross-validation
results showed that the two linear models, BSS and LARS, provided the best
predictions (Supplementary Table 4). The relative importance of the variables
determined by LARS and BSS for predicting the response variables was computed
as the standardized regression coefficient. The higher the coefficient, the higher its
relative importance for the prediction of the SOC response variables.
Testing interrelations of climatic and geochemical predictors.
Bivariate
relationships between climatic variables (precipitation and temperature) and SOC
response variables were estimated using zero-order correlations and partial
correlations by controlling for a single geochemical variable and all geochemical
variables combined. Similarly, relationships with geochemical variables (Al, BS,
clay and silt content, Si and TRB) were controlled for climatic variables.
Partial correlations control the effect of a given predictor on the relationships
between other predictors and response variables, and estimate the strength of the
linear associations between two variables (for example, SOC
Stock
with Si) that
cannot be accounted for by the variability in one or several other variables (for
example, precipitation). A more detailed interpretation and discussion of the
partial correlation results is given in Supplementary Information 1.1.
Level of significance, multicollinearity and model quality assessment.
The
significance of correlations at the 0.05 level was computed using a
t
-test with the
null-hypothesis that there is no correlation. For partial correlation, the test was
adjusted by using (
n
−
2)-
s
degrees of freedom, where
n
is the number of
observations and
s
is the number of variables partialed out. Before model fitting,
high multicollinearity amongst the independent variables was reduced by
computing the variance inflation factor (VIF) and iteratively removing variables
with the highest VIF (sand content, Al/Si ratio), until only variables with VIF
<
10
remained. Linear models were checked for normality and homoscedasticity. Owing
to the relatively low sample size (
n
=
24 plots with three replicates), and to avoid
over-interpreting the data, no interaction terms were allowed to enter the
prediction models and all models were cross-validated. The same cross-validation
partitions were used in all models. For cross-validation, we applied a
leave-one-group-out approach with 20 repetitions/data partition having a 0.75
selection probability. Samples were selected by splitting the dependent variable into
groups based on percentiles and sampling within these groups. Each data partition
was used to fit a model, predict the remaining samples and compute squared errors.
The models were sequentially developed on a grid of model parameters (fraction of
`
1-norm of the standardized coefficients to the maximum
`
1-norm for LARS and
maximum number of variables for BSS regression). To avoid producing complex
CUBIST models, the parameters were fixed and set to zero for the number of
neighbours and one for the number of committees. Model performance was
evaluated with the Root Mean Square Error of Cross-Validation (RMSE
CV
, that is,
mean RMSE over the 20 partitions). The best model parameters were defined as
those producing the most parsimonious model having a RMSE
cv
within one
standard error of the minimal observed RMSE
cv
, as suggested in ref. 36.
References
22. Bannister, J. R., Vidal, O. J., Teneb, E. & Sandoval, V. Latitudinal patterns and
regionalization of plant diversity along a 4270-km gradient in continental
Chile.
Aust. Ecol.
37,
500–509 (2012).
23. Xu, X., Zhou, Y., Ruan, H., Luo, Y. & Wang, J. Temperature sensitivity increases
with soil organic carbon recalcitrance along an elevational gradient in the Wuyi
Mountains, China.
Soil Biol. Biochem.
42,
1811–1815 (2010).
24. Hijmans, R. J., Cameron, S. E., Parra, J. L., Jones, P. G. & Jarvis, A. Very high
resolution interpolated climate surfaces for global land areas.
Int. J. Climatol.
25,
1965–1978 (2005).
25. Mardones, M. in
Historia, Biodiversidad y Ecología de los Bosques Costeros de
Chile
[in Spanish] (eds Smith-Ramirez, C., Armesto, J. & Valdovinos, C.) 39–59
(Editorial Universitaria, 2005).
26. Luebert, F. & Pliscoff, P.
Sinopsis Bioclimática y Vegetacional de Chile
[in
Spanish] (Editorial Universitaria, 2006).
27. Hendershot, W. H., Lalande, H. & Duquette, M. in
Soil Sampling and Methods
of Analysis
(ed. Carter, M. R.) 167–176 (Lewis Publishers, 1993).
28. Bouyoucos, G. J. Hydrometer method improved for making particle size
analysis of soils.
Agron. J.
54,
464–465 (1962).
29. Herbillon, A. J. in
Proc. 8th Int. Soil Classification Workshop
(eds Beinroth, F. H., Camargo, M. N. & Eswaran, H.) 39–48
(EMBRAPA-SNLCS, 1986).
30. Paul, E. A., Morris, S. J. & Boehm, S. in
Assessment Methods for Soil Carbon
(
Advances in Soil Science
)(eds Lal, R., Kimble, J. M., Follett, R. F. &
Stewart, B. A.) 193–206 (CRC/Lewis Publishers, 2001).
31. Stewart, C. E., Paustian, K., Plante, A. F., Conant, R. T. & Six, J. Soil carbon
saturation: Linking concept and measurable carbon pools.
Soil Sci. Soc. Am. J.
72,
379–392 (2008).
32. Zimmermann, M., Leitfeld, J., Abiven, S., Schmidt, M. W. I. & Fuhrer, J. Sodium
hypochlorite separates an older soil organic matter fraction than acid
hydrolysis.
Geoderma
139,
171–179 (2007).
33. Mikutta, R. & Kaiser, K. Organic matter bound to mineral surfaces: Resistance
to chemical and biological oxidation.
Soil Biol. Biochem.
43,
1738–1741 (2011).
34. R Core Team
R: A Language and Environment for Statistical Computing
(R Foundation for Statistical Computing, 2014);
http://www.R-project.org
35. Quinlan, J. R.
Proc. 5th Austral. Join. Con. Art. Intel.
343–348
(World Scientific, 1992).
36. Hastie, T., Tibshirani, R. & Friedman, J.
The Elements of Statistical Learning
(Springer, 2009).
NATURE GEOSCIENCE
|
www.nature.com/naturegeoscience
© 2015 Macmillan Publishers Limited. All rights reserved
Document Outline