Page 1
Article
https://doi.org/10.1038/s41467-024-49743-7
Global turnover of soil mineral-associated
and particulate organic carbon
Zhenghu Zhou
1,2
, Chengjie Ren
3
, Chuankuan Wang
1
,
Manuel Delgado-Baquerizo
4
, Yiqi Luo
5
, Zhongkui Luo
6
, Zhenggang Du
1
,
Biao Zhu
7
, Yuanhe Yang
8
, Shuo Jiao
9
, Fazhu Zhao
10
, Andong Cai
11
,
Gaihe Yang
3
& Gehong Wei
9
Soil organic carbon (SOC) persistence is predominantly governed by mineral
protection, consequently, soil mineral-associated (MAOC) and particulate
organic carbon (POC) turnovers have different impacts on the vulnerability of
SOC to climate change. Here, we generate the global MAOC and POC maps
using 8341 observations and then infer the turnover times of MAOC and POC
by a data-model integration approach. Global MAOC and POC storages are
975
987
964
Pg C (mean with 5% and 95% quantiles) and 330
337
323
Pg C, while global
mean MAOC and POC turnover times are 129
383
45
yr and 23
82
5
yr in the top meter,
respectively. Climate warming-induced acceleration of MAOC and POC
decomposition is greater in subsoil than that in topsoil. Overall, the global atlas
of MAOC and POC turnover, together with the global distributions of MAOC
and POC stocks, provide a benchmark for Earth system models to diagnose
SOC-climate change feedback.
Soil organic carbon (SOC) is the property of terrestrial ecosystems due
to its multi-functionality
1
. Positive feedback to climate change would
occur if SOC is transferred to the atmosphere by a warming-induced
acceleration of its decomposition
2
. Recent advances attest that the
persistence of SOC to microbial decomposition is predominantly
controlled by physical accessibility and isolation rather than its bio-
chemical recalcitrance
3
–
6
. Separating SOC into mineral-associated
(MAOC) and particulate organic carbon (POC) consequently enables
a more accurate prediction of soil vulnerability to climate change
7
,
8
.
Most Earth system models with unmeasurable Century and RothC
carbon pools place low con
fi
dence on both spatial-temporal SOC
distribution and projection to climate change
9
–
11
. Incorporating the
measurable and biophysically de
fi
ned MAOC and POC pools into Earth
system models probably is an optimal approach to trace the fate of
SOC under global change
12
–
15
and would be the next generation of soil
carbon cycling model
16
,
17
.
The critical drivers and global distributions of MAOC and POC are
not well quanti
fi
ed, and the corresponding data product with high
accuracy is also lacking for modeling community. For example, Geor-
giou et al.
18
have created a global map of MAOC but excluding much of
the boreal region with a small dataset (1144 soil observations). Syn-
thesizing 162 soil observations, a recent study suggested that topsoil in
Received: 3 January 2024
Accepted: 18 June 2024
Check for updates
1
College of Ecology and Key Laboratory of Sustainable Forest Ecosystem Management-Ministry of Education, Northeast Forestry University, Harbin, Hei-
longjiang, China.
2
Northeast Asia Biodiversity Research Center, Northeast Forestry University, Harbin, Heilongjiang, China.
3
State Key Laboratory for Crop
Stress Resistance and High-Ef
fi
ciency Production, College of Agronomy, Northwest A&F University, Yangling 712100 Shaanxi, China.
4
Laboratorio de Bio-
diversidad y Funcionamiento Ecosistémico. Instituto de Recursos Naturales y Agrobiología de Sevilla (IRNAS), CSIC, Av. Reina Mercedes, Sevilla, Spain.
5
School of Integrative Plant Science, Cornell University, Ithaca, NY, USA.
6
College of Environmental and Resource Sciences, Zhejiang University,
Hangzhou, China.
7
Institute of Ecology, College of Urban and Environmental Sciences, and Key Laboratory for Earth Surface Processes of the Ministry of
Education, Peking University, Beijing, China.
8
State Key Laboratory of Vegetation and Environmental Change, Institute of Botany, Chinese Academy of
Sciences, Beijing, China.
9
State Key Laboratory for Crop Stress Resistance and High-Ef
fi
ciency Production, College of Life Sciences, Northwest A&F University,
Yangling 712100 Shaanxi, China.
10
Shaanxi Key Laboratory of Earth Surface System and Environmental Carrying Capacity, Northwest University, Xi
’
an,
Shaanxi, China.
11
Institute of Environment and Sustainable Development in Agriculture, Chinese Academy of Agricultural Sciences, Beijing, China.
e-mail:
Rencj1991@nwsuaf.edu.cn
;
weigehong@nwsuaf.edu.cn
Nature Communications
| (2024) 15:5329
1
12
3
4
5
6
78
9
0
()
:,;
12
3
4
5
6
7
8
9
0
()
:,;
Page 2
cold regions was distributed predominantly by more vulnerable POC
rather than by MAOC
8
. MAOC with simple molecular compounds
persists in soils over hundreds of years and is predominantly protected
by minerals via van der Waals force, hydrogen bond, covalent bond,
complexation, cation bridging, and pore sorption
19
,
20
; while POC with
complex molecular compounds persists in soils up to decades relying
on biochemical recalcitrance and aggregate protection
7
,
19
. Therefore,
the global distributions of MAOC and POC may be dependent on dif-
ferent factors. SOC storage capacity can be given as the product of
carbon in
fl
ux and turnover time
21
. Assessing SOC turnover time can
offer an insight into the SOC dynamics under environmental changes
22
,
which is estimated via multiple approaches, such as radiocarbon
23
,
remote sensing
24
,
25
, and Earth-system models
26
. In addition, subsoils
(>20 cm) contain more than half of global SOC stocks
27
. However, the
uncertainty of the response of SOC to climate warming increases with
soil depth because the mechanisms that control deeper SOC
’
s turn-
over are still one of the largest challenges for soil carbon cycling
28
–
30
.
Although MAOC and POC turnovers have very different impacts on soil
fertility and carbon emission
7
, the global distributions of MAOC and
POC turnover times across spatial and soil pro
fi
les, and their covaria-
tions with climate, topography, vegetation, and edaphic factors have
not been quanti
fi
ed.
Here, we synthesized 8341 soil observations that separate the bulk
SOC into MAOC and POC fractions globally (Supplementary
Fig. 1 and 2). The current datasets spanned broadly along spatial and
climatic gradients: latitude ranged from
−
63° S to 69° N, longitude
ranged from
−
155° S to 153° N, mean annual temperature (MAT) ranged
from
−
9 °C to 29 °C, while mean annual precipitation ranged from
58 mm to 3128 mm. We aimed to produce global maps of MAOC and
POC by a machine learning approach, and to reveal the global spatial
and pro
fi
le distributions of turnover times of MAOC and POC using a
data-assimilation approach.
We generate the MAOC and POC maps by soil layers of 0
–
20,
20
–
40, 40
–
60, 60
–
80, and 80
–
100 cm using random forest model
with high accuracy (
R
2
> 0.83). Global MAOC and POC storages are
975
987
964
Pg C (mean with 5% and 95% quantiles from 100-times boot-
strapping) and 330
337
323
Pg C in the top meter, respectively. We also
propose a two-pool model including the measurable MAOC and POC
to infer their turnover times along the soil pro
fi
le. The global mean
MAOC and POC turnover times in the top meter are 129
383
45
yr and 23
82
5
yr, respectively. MAT is the most important factor for MAOC and POC
turnovers and their distribution along the soil pro
fi
le. Climate
warming-induced acceleration of MAOC and POC decomposition is
greater in subsoil than that in topsoil. Our global estimations of the
stocks and turnover times of MAOC and POC would provide a
benchmark for Earth system models to diagnose SOC-climate change
feedback.
Results and Discussion
Global distributions of mineral-associated and particulate
organic carbon
Geographical distributions of MAOC and POC were driven by multiple
climate, topography, vegetation, and edaphic factors (Figs.
1
and
2
;
Supplementary Fig. 3). MAT, mean annual precipitation, net primary
productivity, elevation, pH, clay plus silt content, and sampling depth
were the best predictors of MAOC, which explained 83% variance
(Fig.
1
). Land cover was also important for the spatial variation of POC,
together with the above factors for MAOC, which explained 84% var-
iance for POC (Fig.
1
). As expected, soils with greater carbon inputs
(higher net primary productivity) and lower decomposition rate (lower
MAT) have greater MAOC and POC (Fig.
2
). There is controversy over
whether MAOC is saturated
31
,
32
. The objectors of the concept of MAOC
saturation suggested that organic matter can bind to other organic
matter bonded to minerals, in unlimited
“
skyscraper-like
”
structures
31
.
Here, the partial dependence plot showed that there was a saturation
relationship between MAOC and net primary productivity (carbon
inputs) considering the multiple covariates (Fig.
2
). Therefore, our
global dataset supported the concept of mineral-associated organic
matter saturation.
Both MAOC and POC were higher in acidic soils (Fig.
2
) because
microbial activities are generally inhibited by low pH
33
,
34
. In addition,
soil pH was the second important predictor for the spatial variation of
MAOC, because soil pH mediates many geochemical processes
35
.
Especially, carbon sorption onto soil minerals is pH-dependent, and
mineral surfaces become less positive and organic ligands become
more negative at higher pH, which would minimize the carbon sorp-
tion to mineral surfaces
14
,
20
,
36
–
38
. The mechanisms of SOC stabilization
shift with rising pH from predominantly organo-metal complexation,
Upper depth
Bottom depth
Elevation
MAP
Clay+Silt
pH
NPP
MAT
0.00
0.25
0.50
0.75
1.00
Importance (normalized)
Predictors
MAOC (
R
2
=0.83)
a
Clay+Silt
MAP
Elevation
pH
NPP
Land cover
Upper depth
Bottom depth
MAT
0.00
0.25
0.50
0.75
1.00
Importance (normalized)
Predictors
POC (
R
2
=0.84)
b
Fig. 1 | The importance of predictors for mineral-associated and particulate
organic carbon. a
The importance of predictors for mineral-associated organic
carbon (MAOC).
b
The importance of predictors for particulate organic carbon
(POC). Two types of importance from random forest algorithms (impurity and
permutation) were
fi
rst normalized to the interval of [0, 1] according to their
maximum and minimum values. The mean of the normalized importance was then
calculated. MAT mean annual temperature, MAP mean annual precipitation, NPP
net primary productivity. Source data are provided as a Source Data
fi
le.
Article
https://doi.org/10.1038/s41467-024-49743-7
Nature Communications
| (2024) 15:5329
2
Page 3
to association with short-range-order phases, to Ca complexation and
cation bridging with phyllosilicates, increasing soil pH consequently
decreases the accumulation of MAOC
38
. Soil clay- and silt-size minerals
provide the scaffolds and habitats for MAOC, together with the inha-
bitation of high clay plus silt content on microbial decomposition by
other mechanisms, soil clay plus silt had positive effect on MAOC
(Fig.
2
). In addition, soil minerals can protect POC from microbial
decomposition
39
, and a recent study had provided the visible evidence
that occlusion of organic matter into aggregates and formation of
organo-mineral associations occur concurrently on POC
40
. Land cover
was important for POC but not for MAOC (Fig.
2
and Supplementary
Fig. 3), implying that the quality of litter may have a weak in
fl
uence on
the formation of MAOC at a global scale. In addition, POC may be more
sensitive to the conversion of natural land cover types to croplands
than MAOC.
Finally, the 10-fold validation showed that the random forest had
good performance in predicting the global distributions of MAOC and
POC, with
R
2
of 0.92 and 0.93, respectively (Supplementary Fig. 4). We
generated the MAOC and POC maps by soil layers of 0
–
20, 20
–
40,
40
–
60, 60
–
80, and 80
–
100 cm, respectively (Fig.
3
; Supplementary
Figs. 5 and 6). POC showed a shallower distribution than MAOC
because POC was more dependent upon carbon inputs than soil
properties (Figs.
1
and
3
). We derived a global MAOC storage of 975
987
964
Pg C and POC storage of 330
337
323
Pg C in the top meter (Fig.
3
). Totally,
the global SOC storage is 1306
1321
1292
Pg C, which is well consistent with
previous estimations from observations (ranged from 504 Pg C to 1849
Pg C)
41
and modeled methods (ranged from 510 Pg C to 3040 Pg C)
11
.
Global turnover of mineral-associated and particulate
organic carbon
Using a proposed two-pool model and a data assimilation approach
(see methods), the global mean MAOC and POC turnover times in the
top meter of soils were estimated as 129
383
45
yr and 23
82
5
yr, respectively
(Fig.
4
). MAT is the most important factor for MAOC and POC
M
M
AOC
POC
1.0
1.5
2.0
2.5
−10
0
10
20
30
MAT (
q
C)
ln(MA
O
C or POC) (g
kg
1
)
a
1.0
1.5
2.0
2.5
0
1000
2000
3000
MAP (mm)
ln(MA
O
C or POC) (g
kg
1
)
b
1.0
1.5
2.0
2.5
0.0
0.5
1.0
1.5
2.0
NPP (kg m
2
yr
1
)
ln(MA
OC or POC) (g
kg
1
)
c
0.5
1.0
1.5
2.0
2.5
0
25
50
75
100
125
Upper depth (cm)
ln(MA
O
C or POC) (g kg
1
)
d
0.5
1.0
1.5
2.0
0
50
100
150
Bottom depth(cm)
ln(MA
O
C or POC) (g kg
1
)
e
0.5
1.0
1.5
2.0
2.5
0
25
50
75
100
Clay+Silt (%)
ln(MA
O
C or POC) (g kg
1
)
f
1.2
1.6
2.0
2.4
4
6
8
10
pH
ln(MA
O
C or POC) (g kg
1
)
g
1.2
1.6
2.0
2.4
0
1000
2000
3000
4000
5000
Elevation (m)
ln(MA
O
C or POC) (g kg
1
)
h
0.0
0.5
1.0
1.5
Tundr
a
B
oreal
F
Te
mper
ate F
(S
ub
)T
ropi
ca
l F
Montane GS
Te
mper
at
e
G
S
(Sub
)T
ropica
l GS
Medi
terr
anean
D
eser
ts
Cropl
a
nds
W
et
lands
Land cover
ln(POC) (g kg
1
)
i
Fig. 2 | Partial dependences showing the effects of predictors on mineral-
associated and particulate organic carbon. a
Partial dependences for mean
annual temperature (MAT).
b
Partial dependences for mean annual precipitation
(MAP).
c
Partial dependences for net primary productivity (NPP).
d
Partial depen-
dences for upper sampling depth.
e
Partial dependences for bottom sampling
depth.
f
Partial dependences for clay plus silt content (clay+silt).
g
Partial depen-
dences for pH.
h
Partial dependences for elevation.
i
Partial dependences for land
cover. F forests, GS grasslands and shrublands. Source data are provided as a
Source Data
fi
le.
Article
https://doi.org/10.1038/s41467-024-49743-7
Nature Communications
| (2024) 15:5329
3
Page 4
turnovers (Supplementary Figs. 7 and 8), supporting a previous view-
point that temperature dominates the SOC turnover
22
,
26
. Semi-arid, dry
sub-humid, and humid regions had similar turnover-MAT relation-
ships, but differed from the relationship in arid regions (Supplemen-
tary Fig. 8). Therefore, moisture can dominate the turnover in places
that are highly water-limited, such as the long-term turnover times of
MAOC and POC in desert regions (Fig.
4
).
The turnover times of MAOC and POC were increased as increased
soil depth (Fig.
4
). Referring to the Community Land Model (CLM)
42
, the
reduced rate of decomposition rate with soil depth in our two-pool
model was regulated by an e-folding depth for carbon turnover. A
greater value of e-folding depth for carbon turnover represents a slower
decrease in decomposition rates with increased soil depth, while a
smaller value of e-folding depth for carbon turnover represents a more
rapid one (see methods). MAT predominately controls the variations of
e-folding depths for MAOC and POC turnovers (Fig.
5
). First, cold land
cover (tundra and boral forest) had shallower root distribution than
warm ones (such as (sub)tropical forests and grasslands)
43
. The absence
of fresh carbon is suggested to be a critical mechanism that prevents the
decomposition in deep soil layers because microbial activity was limited
by an essential source of energy
44
,
45
. Second, the slow SOC decom-
position at subsoil could result from a lack of oxygen. Physically, warm
climate would enhance the gaseous diffusivity between subsoil and
atmosphere
46
, subsoils in (sub)tropical regions consequently had lower
oxygen limitation than that in cold regions. We also found that the
e-folding depth for carbon turnover for POC was about 1.7 times as large
as that of MAOC, indicating that an increase in soil depth had stronger
inhibition on MAOC decomposition than that on POC. The total mineral
surface is one of the most limiting factors in
fl
uencing SOC bound to soil
minerals. Compared with topsoil, subsoils had a greater carbon
saturation de
fi
cit, while a greater carbon saturation de
fi
cit not only
increases the stabilization of MAOC but also increases the transfer
ef
fi
ciency from POC to MAOC
47
.
Limitations and uncertainties
There are several limitations and uncertainties in the current study.
First, like the limitation for all global synthesis, the soils were sampled
by different methods, time, successional stage, repeatability, and so
on, which would result in uncertainties. Second, the assumption of
steady
was suggested
to
improve the operability
of
data
assimilation
48
,
49
. Although SOC in ecosystems with anthropogenic
disturbance, such as degraded and restored ecosystems, may not be at
a steady state, previous studies have shown that such a disequilibrium
component is minor for SOC pools considering its long turnover
time
50
,
51
. However, such an assumption of a steady state indeed intro-
duces uncertainties. Third, the data points in Russia, Canada, and
Africa are relatively small, which may limit the accuracy of our esti-
mates of MAOC and POC in these regions, especially for permafrost in
Russia and Canada, which are rich in SOC.
Despite the above limitations, to our knowledge, we presented
the
fi
rst global atlas of MAOC and POC turnover times using a big
dataset of 8341 soil observations, together with the global distribu-
tions of MAOC and POC stocks, which provided a benchmark for
Earth system models. Our
fi
ndings supported that temperature
dominantly controls both storage and turnover times of MAOC and
POC. In addition, the positive effects of MAT on e-folding depths for
both MAOC and POC turnovers imply that climate warming-induced
increase in decomposition was greater in subsoil than that in topsoil.
Therefore, the loss of MAOC and POC in deep soils may have positive
feedback to climate change, it is urgent to prevent the SOC loss from
deep soils.
Methods
Data collection
An extensive literature survey was conducted through the Google
Scholar and China National Knowledge Infrastructure databases until
2024 using keywords of
“
soil
” “
mineral-associated
”
and
“
carbon
”
.
−30
0
30
60
−120
−60
0
60
120
Longitude (
q
)
Latitude (
q
)
10
20
MAOC (kg m
2
)
a
−30
0
30
60
−120
−60
0
60
120
Longitude (
q
)
Latitude (
q
)
2.5
5.0
7.5
10.0
POC (kg m
2
)
b
Modeled
S
p
atiall
y
ex
p
licit
0
500
1000
1500
2000
2500
POC
MAOC
SOC
Previous SOC
Global estimations
SOC (Pg C)
Depth
0−20
20−40
40−60
60−80
80−100
c
Fig. 3 | Global estimations of mineral-associated and particulate organic car-
bon storages. a
Global map of mineral-associated organic carbon (MAOC) storage
(0
–
100 cm).
b
Global map of particulate organic carbon (POC) storage (0
–
100 cm).
c
Global soil organic carbon (SOC) storages from our estimation (bars) and
previous estimations from observations
41
(blue boxplot) and modeled methods
11
(black boxplot). The error bars are 5% and 95% quantiles from 100-times boot-
strapping. Center line, median. Box limits, upper and lower quartiles. Whiskers, 1.5
times interquartile range. Source data are provided as a Source Data
fi
le.
Article
https://doi.org/10.1038/s41467-024-49743-7
Nature Communications
| (2024) 15:5329
4
Page 5
Differing from a recent global synthesis that collected MAOC passed
through 63-
μ
m sieve (1144 entries)
18
, but the sieve size varied greatly
across all the entries, where 53-
μ
m sieve size has the largest proportion
(60.1%), followed by 20-
μ
m sieve size (24.1%) and 63-
μ
m sieve size
(7.8%). To reduce the sieve size impacts from different entries, our
datasets just recorded the MAOC passed through 53-
μ
m sieve. We only
included studies of synchronous reporting SOC, POC, and MAOC.
Finally, our datasets included 8341 entries spanning from 1990 to 2022
(Supplementary Figs. 1 and 2).
To quantify the drivers of global distributions of POC and MAOC,
we consider the effects of climate (MAT and mean annual precipita-
tion), topography (elevation and slope), vegetation (net primary pro-
ductivity and land cover types), soil properties (soil types, pH, clay plus
silt content, base saturation, and cation exchange capacity), nitrogen
deposition, sampling depth and time. These covariates were either
recorded from the original studies or obtained from global datasets
(Supplementary Table 1). The land cover was classi
fi
ed into tundra,
boreal forest, temperate conifer forest, temperate broadleaf/mixed
forest, (sub)tropical moist forest, (sub)tropical dry forest, montane
grassland/shrubland, temperate grassland/shrubland, (sub)tropical
grassland/shrubland, Mediterranean, desert, wetland, and cropland
according to the Terrestrial Ecoregions of the World and the MODIS
Land Cover product. Most of the data entries were from 2000
–
2021
(Supplementary Fig. 2), together with the time range of MODIS net
primary productivity, we calculated the average annual temperature,
precipitation, and net primary productivity from 2001 to 2021.
Therefore, we kept the consistent times for climates, net primary
productivity, MAOC, and POC.
Mapping global mineral-associated and particulate organic car-
bon
. First, we compared the performances of six typical machine
learning algorithms, including random forest, extreme gradient
boosting, support vector machines, recursive partitioning and
regression trees, neural networks, and multivariable linear regression.
We found that the random forest had the least root mean square error
(Supplementary Table 2), therefore, random forest was used to general
the global maps of MAOC and POC.
Second, semivariance was
fi
rst calculated to explore the degree of
spatial auto-correlation of MAOC and POC contents, while the semi-
variograms did not show strong spatial auto-correlation (Supplemen-
tary Fig. 9). In addition, Moran
’
s index was also used to test the
signi
fi
cance of spatial autocorrelation in random forest models. A
greater signi
fi
cantly positive Moran
’
s index indicates that there is
spatial autocorrelation for a given variable at a speci
fi
c distance
threshold. Here, we did not
fi
nd a signi
fi
cant spatial autocorrelation for
MAOC and POC concentrations in random forest models (Supple-
mentary Fig. 9).
Third, the
“
wrapper
”
method
52
was used to select the covariates
among climate (MAT and MAP), topography (elevation and slope),
vegetation (net primary productivity and land cover types), soil
properties (soil types, pH, clay plus silt content, base saturation, and
cation exchange capacity), nitrogen deposition, sampling depth and
time. Speci
fi
cally, by re-calibrating a random forest model several
times, each time removing the least important covariate, one may
expect to considerably reduce the overall number of covariates with
little or no decrease in model prediction accuracy. We used methods
of impurity and permutation in the ranger function of the
“
ranger
”
package to quantify the importance of predictors, we then normal-
ized the two types of importance to the interval of [0, 1] according to
their maximum and minimum values. The mean of the normalized
importance was used. The
fi
nal random forest model for MAOC
included MAT, mean annual precipitation, net primary productivity,
−30
0
30
60
−120
−60
0
60
120
Longitude (
q
)
Latitude (
q
)
10
100
1000
W
MAOC
(yr)
a
−30
0
30
60
−120
−60
0
60
120
Longitude (
q
)
Latitude (
q
)
10
50
300
W
POC
(yr)
b
129
1
45
383
23
5
82
41
14
106
13
3
39
69
24
186
18
4
61
116
40
329
26
6
95
198
68
588
38
8
152
340
109
1064
58
10
252
80−100
60−80
40−60
20−40
0−20
0−100
10
100
1000
Turnover time (yr)
Depth (cm)
MAOC
POC
c
Fig. 4 | Spatial and pro
fi
le patterns of turnover times of mineral-associated and
particulate organic carbon. a
mineral-associated organic carbon (MAOC) turn-
over time (0
–
100 cm).
b
particulate organic carbon (POC) turnover time
(0
–
100 cm).
c
Pro
fi
le patterns of turnover times of MAOC and POC. The error bars
are 5% and 95% quantiles from 100-times bootstrapping. Source data are provided
as a Source Data
fi
le.
Article
https://doi.org/10.1038/s41467-024-49743-7
Nature Communications
| (2024) 15:5329
5
Page 6
elevation, pH, clay plus silt content, and sampling depth, while the
fi
nal random forest model for POC included MAT, mean annual
precipitation, net primary productivity, land cover, elevation, pH,
clay plus silt content, and sampling depth (Fig.
1
and Supplemen-
tary Fig. 3).
Furthermore,
K
-fold cross-validation (with
K
= 10) was used to
validate the random forest models, and our random forest model
showed good performance (Supplementary Fig. 4). To generate the
spatially-explicit and quantitative maps of MAOC and POC contents,
the global predictors were aggregated or resampled at a resolution of
0.5° × 0.5° by soil layers (0
–
20, 20
–
40, 40
–
60, 60
–
80, and
80
–
100 cm). The 100-time bootstrapping was used to generate the
uncertainties of global maps (for each bootstrapping, 90% of the data
were randomly sampled).
Model desperation
. A two-pool model of MAOC and POC was pro-
posed following the framework Century (Supplementary Fig. 10):
d
POC
i
ð
t
Þ
d
t
=
I
i
+
MAOC
i
KM
i
T
MAOCtoPOC
POC
i
KP
i
+
POC
i
+ 1
D
POC
i
A
ð
1
Þ
Soil types
Elevation
Slope
Clay+Silt
N deposition
CEC
MAP
pH
Land cover
Bast
Bulk density
MAT
0.00
0.25
0.50
0.75
1.00
Importance (normalized)
Predictors
a
Soil types
Bast
pH
Slope
Clay+Silt
MAP
Bulk density
Elevation
CEC
N deposition
Land cover
MAT
0.00
0.25
0.50
0.75
Importance (normalized)
Predictors
b
t
=−90.2,
P
<0.001
50
75
100
MAOC
POC
Fractions
The e−f
olding depth f
o
r C tur
n
o
v
er
c
38
39
40
41
42
−10
0
10
20
30
MAT (
q
C)
The e−f
olding depth f
o
r MA
OC tur
n
o
v
er
d
66
68
70
72
−10
0
10
20
30
MAT (
q
C)
The e−f
olding depth f
o
r POC tur
n
o
v
er
e
Fig. 5 | Factors in
fl
uencing the reduced rate of decomposition rate with
soil depth. a
and
b
Importance of predictors for e-folding depths for mineral-
associated (MAOC) and particulate organic carbon (POC) turnovers, respectively. A
greater value of e-folding depth for carbon turnover represents a slower reduced
rate of decomposition rate with increased soil depth, while a smaller value of
e-folding depth for carbon turnover represents a more rapid one. Two types of
importance (impurity and permutation) were
fi
rst normalized to the interval of [0,
1] according to their maximum and minimum values. The mean of the normalized
importance was then calculated.
c
The e-folding depths for MAOC vs. POC turn-
overs (mean with 5% and 95% quantiles) with paired
t
-test.
d
and
e
Partial depen-
dence showing the effect of mean annual temperature (MAT) on e-folding depths
for MAOC and POC turnovers, respectively. MAP mean annual precipitation. NPP
net primary productivity. CEC cation exchange capacity. Bast base saturation.
Source data are provided as a Source Data
fi
le.
Article
https://doi.org/10.1038/s41467-024-49743-7
Nature Communications
| (2024) 15:5329
6
Page 7
d
MAOC
i
ð
t
Þ
d
t
=
POC
i
KP
i
T
POCtoMAOC
MAOC
i
KM
i
+
MAOC
i
+ 1
D
MAOC
i
A
ð
2
Þ
where
i
is
i
th soil depth, referring to 0
–
20, 20
–
40, 40
–
60, 60
–
80, and
80
–
100 cm, respectively.
I
i
is net primary productivity allocation into
i
th soil depth, which refers to a recent global study that has quanti
fi
ed
the allocation of net primary productivity into belowground by the
same soil depth
43
. Therefore, we
fi
rstly obtained the averaged MODIS
net primary productivity from 2001 to 2023. We then calculated the
net primary productivity allocation into 0
–
20, 20
–
40, 40
–
60, 60
–
80,
and 80
–
100 cm, respectively.
MAOC
i
and
POC
i
are MAOC and POC
pools of
i
th soil depth.
KM
i
and
KP
i
are the decomposition rates
(inverse of turnover time) of MAOC and POC, respectively.
T
MAOCtoPOC
is SOC transformation from MAOC to POC, while
T
POCtoMAOC
is SOC
transformation from POC to MAOC.
D
and
A
represent the diffusion
and advection, respectively.
Referring to Community Land Model (CLM)
42
, an exponential
decrease in decomposition was used to account for the regulation of
soil depth on decomposition:
KM
i
=
KM
1
exp
z
i
10
z
M
ð
3
Þ
KP
i
=
KP
1
exp
z
i
10
z
P
ð
4
Þ
where
z
M
and
z
P
are the e-folding depths for carbon turnovers for
MAOC and POC, respectively. z
i
is the soil depth of
i
th layer.
Data assimilation
. We have more robust global estimates of pro-
ductivity than of heterotrophic respiration. Therefore, if we assume
that soils are approximately at a steady state, we can analytically cal-
culate the steady-state pool sizes of MAOC and POC by letting
Eqs. (
1
,
2
) equal zero, i.e., a semi-analytical solution to accelerate spin-
up
53
. Bayesian probabilistic inversion was used to calibrate the para-
meters in the two-pools model
9
,
54
. In Bayes
’
theorem, the posterior
probability density functions (
P
(
θ
|
Z
)) of model parameters (
θ
) can be
obtained from prior probability density functions of the parameters
(
P
(
θ
)) and the likelihood function (
P
(
Z
|
θ
)):
P
ð
θ
j
Z
Þ /
P
ð
Z
j
θ
Þ
P
ð
θ
Þ
ð
5
Þ
The
P
(
Z
|
θ
) was calculated with the assumption that errors
between observed and modeled C pools were independent of each
other and followed a multivariate Gaussian distribution with a zero
mean:
P
ð
Z
j
θ
Þ /
exp
C
obs
C
mod
2
2
σ
2
(
)
ð
6
Þ
where C
obs
and C
mod
are the observed and modeled carbon pools of
soil layers of 0
–
20, 20
–
40, 40
–
60, 60
–
80, and 80
–
100 cm, while
σ
is
the standard deviation of observation.
To generate the posterior distributions of model parameters, we
fi
rst speci
fi
ed the prior ranges of the model parameters in Supple-
mentary Table 3 based on the literature, educated guess, and hypoth-
esis testing
54
. Speci
fi
cally, a previous opinion study proposed that POC
and MAOC have vague turnover times of <10 years to decades and
decades to centuries, respectively. Here, the wide ranges of turnover
times were set (Supplementary Table 3). The e-folding depths were set
as 50 cm in CLM referring to one study of
14
C pro
fi
les
42
, therefore, we
set a range of 30
–
200 cm. The current vertically resolved models, such
as CLM
42
and ORCHIDEE-SOM
55
, simulate the vertical transformation of
SOC using two processes of diffusion and advection. However, these
two
fl
uxes were not well quanti
fi
ed in
fi
elds with diverse conditions. We
proposed a relatively wide of range for these two processes by referring
to CLM
42
and ORCHIDEE-SOM
55
. Finally, according to the global dis-
tributions of diffusion and advection optimized by the following pro-
cesses, our original set was acceptable for diffusion and advection (not
clustered to the upper limit; Supplementary Fig. 11).
Once we speci
fi
ed the parameter ranges, the probabilistic inver-
sion was carried on using the Metropolise-Hastings algorithm, a Mar-
kov Chain Monte Carlo technique, to construct the
P
(
θ
|
Z
)
56
. Brie
fl
y, the
Metropolise-Hastings algorithm includes a proposing step and a
moving step. In the proposing step, a new parameter set
θ
new
is pro-
posed based on the previously accepted parameter set
θ
old
and a
proposal distribution:
θ
new
=
θ
old
+
r
×
ð
θ
max
θ
min
Þ
D
ð
7
Þ
where
θ
max
and
θ
min
are the maximum and minimum values of para-
meters,
r
is a random variable between -0.5 and 0.5, and D is used to
control the proposed step size and was set to 5
56
. In the moving step,
θ
new
was tested against the Metropolis criterion to examine if the new
parameter set should be accepted or rejected. The
fi
rst half of the
accepted samples were discarded and only the rest were used to
generate
P
(
θ
|
Z
), i.e., these optimized parameters had the greatest
accuracy in predicting the MAOC and POC dynamics (Supplemen-
tary Fig. 10).
The turnover times of MAOC (
τ
MAOC
i
) and POC (
τ
POC
i
) in
i
th layers
are the reciprocal of corresponding decay constants:
τ
MAOC
i
=
1
KM
i
ð
8
Þ
τ
POC
i
=
1
KP
i
ð
9
Þ
The two-pool model was conducted with a spatial resolution of
1.5° × 1°, i.e., we optimized model parameters for each grid by obser-
ving pro
fi
le distributions of MAOC and POC. The turnover time of the
whole pro
fi
le (0
–
100 cm) can be calculated as the weighted mean of
turnover time by pool sizes of different soil layers. We also used ran-
dom forests to explore the regulation of climate, topography, vege-
tation, soil properties, and nitrogen deposition on turnover times and
the e-folding depth of intrinsic decomposition rates.
All the statistical analyses were performed in R 4.2.2 (R Core
Team 2022).
Data availability
The code that supports the
fi
ndings of this study are openly available in
fi
gshare at
https://
fi
gshare.com/account/articles/25962769
. Source
data are provided with this paper.
Code availability
The code that supports the
fi
ndings of this study are openly available in
fi
gshare at
https://
fi
gshare.com/account/articles/25962721
.
References
1.
Schmidt, M. W. et al. Persistence of soil organic matter as an eco-
system property.
Nature
478
, 49
–
56 (2011).
2.
Davidson, E. A. & Janssens, I. A. Temperature sensitivity of soil
carbon decomposition and feedbacks to climate change.
Nature
440
, 165
–
173 (2006).
3.
Bailey, V. L. et al. Soil carbon cycling proxies: Understanding their
critical role in predicting climate change feedbacks.
Glob. Change
Biol.
24
, 895
–
905 (2018).
Article
https://doi.org/10.1038/s41467-024-49743-7
Nature Communications
| (2024) 15:5329
7
Page 8
4.
Cotrufo, M. F., Wallenstein, M. D., Boot, C. M., Denef, K. & Paul, E.
The Microbial Ef
fi
ciency
‐
Matrix Stabilization (MEMS) framework
integrates plant litter decomposition with soil organic matter sta-
bilization: do labile plant inputs form stable soil organic matter?
Glob. Change Biol.
19
, 988
–
995 (2013).
5.
Dungait, J. A., Hopkins, D. W., Gregory, A. S. & Whitmore, A. P. Soil
organic matter turnover is governed by accessibility not recalci-
trance.
Glob. Change Biol.
18
, 1781
–
1796 (2012).
6.
Kleber, M. et al. Old and stable soil organic matter is not necessarily
chemically recalcitrant: implications for modeling concepts and
temperature sensitivity.
Glob. Change Biol.
17
, 1097
–
1107 (2011).
7.
Lavallee, J. M., Soong, J. L. & Cotrufo, M. F. Conceptualizing soil
organic matter into particulate and mineral
‐
associated forms to
address global change in the 21st century.
Glob. Change Biol.
26
,
261
–
273 (2020).
8.
García-Palacios, P. et al. Dominance of particulate organic carbon in
top mineral soils in cold regions.
Nat. Geosci.
17
, 145
–
150 (2024).
9.
Shi, Z., Crowell, S., Luo, Y. & Moore, B. Model structures amplify
uncertainty in predicted soil carbon responses to climate change.
Nat. Commun.
9
, 2171 (2018).
10. Sulman, B. N. et al. Multiple models and experiments underscore
large uncertainty in soil carbon dynamics.
Biogeochemistry
141
,
109
–
123 (2018).
11.
Todd-Brown, K. E. et al. Causes of variation in soil carbon simula-
tions from CMIP5 Earth system models and comparison with
observations.
Biogeosciences
10
, 1717
–
1736 (2013).
12. Sokol, N. W., Sanderman, J. & Bradford, M. A. Pathways of mineral
‐
associated soil organic matter formation: Integrating the role of
plant carbon source, chemistry, and point of entry.
Glob. Change
Biol.
25
, 12
–
24 (2019).
13. Liu, F. et al. Divergent changes in particulate and mineral-
associated organic carbon upon permafrost thaw.
Nat. Commun.
13
, 5073 (2022).
14. Abramoff, R. Z. et al. Improved global-scale predictions of soil
carbon stocks with Millennial Version 2.
Soil Biol. Biochem.
164
,
108466 (2022).
15. Robertson, A. D. et al. Unifying soil organic matter formation and
persistence frameworks: the MEMS model.
Biogeosciences
16
,
1225
–
1248 (2019).
16. Opfergelt, S. The next generation of climate model should account
for the evolution of mineral-organic interactions with permafrost
thaw.
Environ. Res. Lett.
15
, 1748
–
9326 (2020).
17. Doetterl, S. et al. Soil carbon storage controlled by interactions
between geochemistry and climate.
Nat. Geosci.
8
, 780
–
783 (2015).
18. Georgiou, K. et al. Global stocks and capacity of mineral-associated
soil organic carbon.
Nat. Commun.
13
, 3797 (2022).
19. Lützow, M. V. 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).
20. Kleber, M. et al. Dynamic interactions at the mineral
–
organic matter
interface.
Nat. Rev. Earth Environ.
2
, 402
–
421 (2021).
21. Xia, J., Luo, Y., Wang, Y. P. & Hararuk, O. Traceable components of
terrestrial carbon storage capacity in biogeochemical models.
Glob. Change Biol.
19
, 2104
–
2116 (2013).
22. Luo, Z., Wang, G. & Wang, E. Global subsoil organic carbon turnover
times dominantly controlled by soil properties rather than climate.
Nat. Commun.
10
, 3688 (2019).
23. Shi, Z. et al. The age distribution of global soil carbon inferred from
radiocarbon measurements.
Nat. Geosci.
13
, 555
–
559 (2020).
24. Carvalhais, N. et al. Global covariation of carbon turnover times with
climate in terrestrial ecosystems.
Nature
514
, 213
–
217 (2014).
25. Bloom, A. A., Exbrayat, J. F., Van Der Velde, I. R., Feng, L. & Williams,
M. The decadal state of the terrestrial carbon cycle: Global retrie-
vals of terrestrial carbon allocation, pools, and residence times.
Proc. Natl Acad. Sci.
113
, 1285
–
1290 (2016).
26. Koven, C. D., Hugelius, G., Lawrence, D. M. & Wieder, W. R. Higher
climatological temperature sensitivity of soil carbon in cold than
warm climates.
Nat. Clim. Change
7
, 817
–
822 (2017).
27. Jobbágy, E. G. & Jackson, R. B. The vertical distribution of soil
organic carbon and its relation to climate and vegetation.
Ecol.
Appl.
10
, 423
–
436 (2000).
28. Zhang, Q. et al. Whole-soil-pro
fi
le warming does not change
microbial carbon use ef
fi
ciency in surface and deep soils.
Proc. Natl
Acad. Sci.
120
, e2302190120 (2023).
29. Wang, M. et al. Global soil pro
fi
les indicate depth-dependent soil
carbon losses under a warmer climate.
Nat. Commun.
13
, 5514
(2022).
30. Hicks Pries, C. E., Castanha, C., Porras, R. & Torn, M. The whole-soil
carbon
fl
ux in response to warming.
Science
355
, 1420
–
1423 (2017).
31. Begill, N., Don, A. & Poeplau, C. No detectable upper limit of
mineral-associated organic carbon in temperate agricultural soils.
Glob. Change Biol.
29
, 4662
–
4669 (2023).
32. Cotrufo, M. F., Lavallee, J. M., Six, J. & Lugato, E. The robust concept
of mineral-associated organic matter saturation: A letter to Begill
et al., 2023.
Glob. Change Biol.
29
, 5986
–
5987 (2023).
33. Averill, C. & Waring, B. Nitrogen limitation of decomposition and
decay: How can it occur?
Glob. Change Biol.
24
, 1417
–
1427 (2018).
34. Lugato, E., Lavallee, J. M., Haddix, M. L., Panagos, P. & Cotrufo, M. F.
Different climate sensitivity of particulate and mineral-associated
soil organic matter.
Nat. Geosci.
14
, 295
–
300 (2021).
35. Slessarev, E. et al. Water balance creates a threshold in soil pH at
the global scale.
Nature
540
, 567
–
569 (2016).
36. Mayes, M. A., Heal, K. R., Brandt, C. C., Phillips, J. R. & Jardine, P. M.
Relation between soil order and sorption of dissolved organic car-
bon in temperate subsoils.
Soil Sci. Soc. Am. J.
76
, 1027
–
1037
(2012).
37. Feng, X., Simpson, A. J. & Simpson, M. J. Chemical and miner-
alogical controls on humic acid sorption to clay mineral surfaces.
Org. Geochem.
36
, 1553
–
1566 (2005).
38. Rasmussen, C. et al. Beyond clay: towards an improved set of
variables for predicting soil organic matter content.
Biogeochem-
istry
137
, 297
–
306 (2018).
39. Lei, L., Holásková, I., Thompson, J. A. & McDonald, L. M. Fe-bound
organic carbon and sorption of aromatic dissolved organic carbon
in surface soil: comparing a forest, a cropland, and a pasture soil in
the Central Appalachian Region, West Virginia, USA.
Environments
9
, 113 (2022).
40. Witzgall, K. et al. Particulate organic matter as a functional soil
component for persistent soil organic carbon.
Nat. Commun.
12
,
4115 (2021).
41. Scharlemann, J. P., Tanner, E. V., Hiederer, R. & Kapos, V. Global soil
carbon: understanding and managing the largest terrestrial carbon
pool.
Carbon Manag.
5
, 81
–
91 (2014).
42. Koven, C. D. et al. The effect of vertically resolved soil bio-
geochemistry and alternate soil C and N models on C dynamics of
CLM4.
Biogeosciences
10
, 7109
–
7131 (2013).
43. Xiao, L. et al. Global depth distribution of belowground net primary
productivity and its drivers.
Glob. Ecol. Biogeogr.
32
, 1435
–
1451
(2023).
44. Fontaine, S. et al. Stability of organic carbon in deep soil layers
controlled by fresh carbon supply.
Nature
450
, 277
–
U210 (2007).
45. Bernal, B. et al. Limits to soil carbon stability; Deep, ancient soil
carbon decomposition stimulated by new labile organic inputs.
Soil
Biol. Biochem.
98
, 85
–
94 (2016).
46. Tang, J. W., Baldocchi, D. D., Qi, Y. & Xu, L. K. Assessing soil CO
2
ef
fl
ux using continuous measurements of CO
2
pro
fi
les in soils with
small solid-state sensors.
Agric. For. Meteorol.
118
, 207
–
220 (2003).
47. Stewart, C. E., Paustian, K., Conant, R. T., Plante, A. F. & Six, J. Soil
carbon saturation: Evaluation and corroboration by long-term
incubations.
Soil Biol. Biochem.
40
, 1741
–
1750 (2008).
Article
https://doi.org/10.1038/s41467-024-49743-7
Nature Communications
| (2024) 15:5329
8
Page 9
48. Zhou, T., Shi, P., Jia, G. & Luo, Y. Nonsteady state carbon seques-
tration in forest ecosystems of China estimated by data assimilation.
J. Geophys. Res.: Biogeosci.
118
, 1369
–
1384 (2013).
49. Zhou, T. et al. Age
‐
dependent forest carbon sink: Estimation via
inverse modeling.
J. Geophys. Res.: Biogeosci.
120
, 2473
–
2492
(2015).
50. Lu, X., Wang, Y.-P., Luo, Y. & Jiang, L. Ecosystem carbon transit
versus turnover times in response to climate warming and rising
atmospheric CO
2
concentration.
Biogeosciences
15
, 6559
–
6572
(2018).
51. Tao, F. et al. Microbial carbon use ef
fi
ciency promotes global soil
carbon storage.
Nature
618
, 981
–
985 (2023).
52. Wadoux, A. M. C., Minasny, B. & McBratney, A. B. Machine learning
for digital soil mapping: Applications, challenges and suggested
solutions.
Earth-Sci. Rev.
210
, 103359 (2020).
53. Xia, J., Luo, Y., Wang, Y. P., Weng, E. & Hararuk, O. A semi-analytical
solution to accelerate spin-up of a coupled carbon and nitrogen
land model to steady state.
Geosci. Model Dev.
5
, 1259
–
1271 (2012).
54. Hararuk, O., Xia, J. & Luo, Y. Evaluation and improvement of a global
land model against soil carbon data using a Bayesian Markov chain
Monte Carlo method.
J. Geophys. Res.: Biogeosci.
119
,
403
–
417 (2014).
55. Camino-Serrano, M. et al. ORCHIDEE-SOM: modeling soil organic
carbon (SOC) and dissolved organic carbon (DOC) dynamics along
vertical soil pro
fi
les in Europe.
Geosci. Model Dev.
11
, 937
–
957 (2018).
56. Xu, T., White, L., Hui, D. & Luo, Y. Probabilistic inversion of a ter-
restrial ecosystem model: Analysis of uncertainty in parameter
estimation and model prediction.
Glob. Biogeochem. Cycles
20
,
GB2007 (2006).
Acknowledgements
We thank Zheng Shi and Yuanyuan Huang for their suggestions on this
study. We thank all the researchers whose data were used in this global
synthesis. This work was
fi
nancially supported by the National Key
Research and Development Program of China (2021YFD2200401), the
National Natural Science Foundation of China (42377345), and the
Fundamental Research Funds for the Central Universities
(2572021CG07).
Author contributions
C.R., Z.Z. and G.W. conceived and designed the project; C.R., S. J., A. C.,
Y. Y. and F. Z. collected the data; Z.Z., B. Z., M. D. and C.R. conducted the
data analyses with help from Y.L., Z. D., G.Y., C.W. and Z.K.; Z.Z. pro-
posed the new two-pool model; Z.Z. and C.R. wrote the manuscript; All
authors contributed to discussing the results, writing, and editing
the paper.
Competing interests
The authors declare no competing interests.
Additional information
Supplementary information
The online version contains
supplementary material available at
https://doi.org/10.1038/s41467-024-49743-7
.
Correspondence
and requests for materials should be addressed to
Chengjie Ren or Gehong Wei.
Peer review information
Nature Communications
thanks the anon-
ymous reviewers for their contribution to the peer review of this work. A
peer review
fi
le is available.
Reprints and permissions information
is available at
http://www.nature.com/reprints
Publisher
’
s note
Springer Nature remains neutral with regard to jur-
isdictional claims in published maps and institutional af
fi
liations.
Open Access
This article is licensed under a Creative Commons
Attribution 4.0 International License, which permits use, sharing,
adaptation, distribution and reproduction in any medium or format, as
long as you give appropriate credit to the original author(s) and the
source, provide a link to the Creative Commons licence, and indicate if
changes were made. The images or other third party material in this
article are included in the article
’
s Creative Commons licence, unless
indicated otherwise in a credit line to the material. If material is not
included in the article
’
s Creative Commons licence and your intended
use is not permitted by statutory regulation or exceeds the permitted
use, you will need to obtain permission directly from the copyright
holder. To view a copy of this licence, visit
http://creativecommons.org/
licenses/by/4.0/
.
© The Author(s) 2024
Article
https://doi.org/10.1038/s41467-024-49743-7
Nature Communications
| (2024) 15:5329
9
Document Outline