Introduction
Study of the long-term variability in atmospheric composition from air
trapped in polar ice has improved our understanding of processes and
feedbacks between climate and the carbon cycle on decadal to millennial
scales and allows us to evaluate the magnitude of human impact on the
Earth's atmosphere. Since the mid-20th century, atmospheric CO2
monitoring has progressed from the first consistent measurements in 1957 on
Mauna Loa to a global network of monitoring sites
and the recent satellite missions to measure CO2 in
the atmospheric column . Direct
measurements of the difference between the partial pressure gradient of
CO2 between sea water and the overlying air (ΔpCO2) have
also been available since the early 1970s . These
measurements enabled the study of variability in sinks and sources of
CO2 at seasonal to interannual timescales. However, most of these
time series are still short, hampering the study of variability on scales
longer than a few decades. As direct high-precision CO2 measurements are
only available for the later decades of the 20th century, ice-core records
remain a valuable source of information about atmospheric CO2
variability and trends during earlier periods.
Atmospheric CO2 concentration in the Law Dome ice core and firn record from and
respective uncertainties (markers and whiskers) as well as the spline fit applied to the data following ,
which attenuates by 50 % variations of ca. 23 years. The period corresponding to the plateau is highlighted between vertical
grey lines. The blue markers correspond to samples from Law Dome and red markers from South Pole; different symbols indicate
the different ice cores (big markers) and firn samples (dots).
The high-resolution measurements of CO2 and the isotopic signature of
carbon (usually expressed as δ13C) in air from ice core and firn
(unconsolidated snow) air from the Law Dome ice sheet in Antarctica
encompass the last
millennium, while extending to the present and overlapping with direct
atmospheric observations. The Law Dome data remain unique for the period of
the 1940s as the other recently drilled high-resolution ice-core CO2
record (from West Antarctic Ice Sheet Divide) is restricted to years
before 1940 due to contamination of gas records in ice collected from shallow
depths . The Law Dome record was used to study the variability
at decadal scales in CO2 sources and sinks during the 20th century
. A conspicuous feature in the
atmospheric CO2 record is a stabilization of CO2 concentration at
around 310–312 ppm from 1940 to the early 1950s (,
Fig. ). This CO2 stabilization was reconfirmed in the
high-density measurements from and more recently
by .
Assuming the estimates and their uncertainty range by are
correct, the CO2 plateau could not have been the result of CO2
emissions from fossil-fuel burning and cement production going down to 0
or decreasing abruptly (which could result in a temporary strong sink, as
discussed in ). The only available land-use change (LUC)
emission estimates over the 20th century, based on observation-driven
(bookkeeping) models, also do not report any decrease of emissions during
this period . , using an
intermediate complexity Earth system model (ESM), found that, even in the absence
of land-use emissions, the atmospheric CO2 concentration should have
risen by 2.6 ppm during the 1940 to 1950 period, a rate comparable to
previous decades. Accounting for LUC scenarios,
and added 0.7–1 ppm to this CO2
rise due to fossil-fuel emissions. Simulated δ13CO2 in the
atmosphere from was in line with the Law Dome record
during 1940s but offset by -0.2 ‰ after 1950, which might
indicate overestimation of the land-use emissions after the Second World War.
Due to the smoothing of the short-term variations of atmospheric CO2,
the plateau in the ice-core record may be either due to a remarkably strong
uptake of CO2 lasting a few years or a sustained uptake matching the
anthropogenic emissions during the period, either by the land or ocean
reservoirs (or a combination of both).
Double deconvolution of CO2 and δ13C suggests that this
increased sink was dominated by ocean uptake
. noticed that
persistent El Niño sequences in 1895–1898 and 1911–1916 as well as the
1940s coincided with small decreases in the CO2 growth rate, and
hypothesized that the very strong El Niño event that
lasted from 1940 until 1942 may have been responsible
for reduced upwelling of carbon-rich waters in the Eastern Pacific, causing
an abnormal increase of the global ocean sink. However, this hypothesis
remains controversial and, moreover, in spite of the high quality of the Law
Dome δ13C record, the scatter and uncertainty in the data are
relatively high and they affect how well it is possible to partition the
biospheric and oceanic fluxes. Errors in the δ13C data may lead to
spurious and highly correlated terrestrial and oceanic fluxes
. The magnitude of uncertainties in the δ13C
ice-core measurements can be hard to estimate accurately, and the choice of
the uncertainty range may result in significant differences in the magnitude
of the resulting fluxes.
, using a single deconvolution of the CO2 record and
a simple land-surface model, pointed to an increased terrestrial sink during
the 1940s. This sink was related to change in temperature. Single
deconvolutions do not use the δ13C information and assume
time-invariant ocean response. When terrestrial uptake is used to explain the
1940s plateau they produce a peak in δ13C that appears to be
inconsistent with the ice-core δ13C measurements, although the
differences are not large compared to the measurement uncertainties
.
Furthermore, even if the unusually long 1940–1942 El Niño did induce strong
oceanic uptake, it is not clear that it should have led to a decrease in
CO2 growth rate, as El Niño periods in recent decades have usually
been associated with a net increase in atmospheric CO2 growth rate
. The occurrence of El Niño leads to
reduced outgassing of CO2 in the tropical Pacific due to the slow-down
of vertical upwelling of carbon and nutrient-rich subsurface waters, driven
by weaker trade winds . However, the magnitude of the El Niño/Southern Oscillation
(ENSO)
impact on oceanic uptake differs significantly between studies, with
approaches based on δ13C analysis pointing to anomalies of
1.5–2.5 Pg C yr-1 , while
atmospheric CO2-based methods point to anomalies of only
0.1–0.5 Pg C yr-1 ,
consistent with the values obtained from observations of ΔpCO2 in
the equatorial Pacific by or for the global ocean
. Furthermore, the enhancement of the global ocean sink
during an El Niño event is usually offset by a much larger terrestrial
CO2 release due to the response of land ecosystems to widespread
drought conditions in the tropics and
increased fire emissions .
Here, we evaluate whether it is possible to reproduce the stabilization in
atmospheric CO2 during the 1940s using model-based records of sources
and sinks for the 20th century and identify possible mechanisms to explain
the plateau. We first compare the atmospheric CO2 growth rate
reconstructed using these datasets with the ice-core record to test their
ability to capture the plateau. Additionally, we evaluate whether the ocean
response to the 1940–1942 El Niño may explain the atmospheric CO2
stabilization. Finally, we analyse the response of the land sink to this
event using land-surface process models and, given that land-use data are
highly uncertain, test the possible contribution of LUC to explain the
additional sink required to match observations.
Methods
Atmospheric CO2
The Law Dome ice-core and firn air records of atmospheric composition were
constructed by analysis of air trapped in impermeable ice cores or in firn
layers at four sites on Law Dome in Antarctica (DE08, DE08-2, DSS, and DSSW20K).
They extend back about 2000 years before present with very high air-age
resolution and measurement precision. It is the only CO2 record covering
the 1940s and 1950s period and overlapping with the start of direct
measurements .
Here, we use the atmospheric CO2 concentration data from air samples
from the DE08, DE08-2, and DSS ice cores and from firn air from the DSSW20K
record and from the South Pole from
, shown in Fig. . To compile this dataset,
added new CO2 and δ13C measurements to the
record and revised sampling methods, uncertainty estimates, and gravitational
and diffusive mixing corrections relative to older measurements (e.g.
).
The enclosure of air pores in ice is a gradual process that occurs in the
lock-in zone, the transition layer between firn and impermeable ice. Due to
the porosity of firn, there is also mixing between different parcels of air
in the overlying layers. Therefore, the composition of air in ice-core
samples corresponds not to a single discrete year in the past but rather
to a mix of air parcels with different ages.
The air-age distribution, i.e. the temporal range of real world atmospheric
composition sampled by each single ice-core measurement, can be quantified by
a model of the processes . Law Dome ice has an average
spectral width (a measure of the spread of the distribution;
) of 4.5 years in DE08 and DE08-2, 5.8 years in DSS ice
samples, and 7.0 years in DSSW20K firn samples
. More information about the characteristics
of the original dataset may be found in .
In order to derive a continuous time series of annual values from the
individual measurements, we fit a smoothing spline curve to the ice-core
measurements following the procedure described by which
allows estimation of uncertainties in the spline, as well as in its
derivative. The derivative corresponds to the annual atmospheric CO2
growth rate (hereafter AGR; Fig. ).
However, when fitting a spline to data, a set of parameters needs to be
chosen (regularization parameter and smoothing weights), which may affect the
resulting spline . We performed sensitivity tests on the
choice of these parameters (Fig. S1 in the Supplement). While different choices of parameters
and weights lead to very similar values of atmospheric CO2 during the
20th century (Fig. S1, top), the CO2 growth rate has much higher
sensitivity to the different choices (Fig. S1, bottom). Here we use
λ = 30, which results in 50 % attenuation of variations shorter than 23 years, comparable to the 20 years of , and use
unit weights for the fit, as in the standard definition of smoothing spline
, although the latter result in higher uncertainty relative
to other choices of weights for the fit.
A running piecewise trend adjustment was performed on the spline data between
1930 and 1960 (Fig. S2) to identify break points, and the existence of a
monotonically increasing or decreasing trend of atmospheric CO2 for each
trend section is tested by a Mann–Kendall test. The fit with the smallest
root mean square error of the adjustment was selected and the corresponding
breakpoints defined as limits of the plateau. The resulting period spans from
1940 to 1950, as highlighted in Fig. by the two vertical
lines. During this period we found no significant trend in atmospheric
CO2. Note that the ice-core record is a smoothed and slightly shifted
representation of the real atmospheric variations, and therefore the sink
anomaly is more likely to be expected a few years after the stabilization
becomes evident in the ice-core record . Previous
analysis of the plateau has varied by a few years in the timing of the
maximum uptake. predicted maximum uptake in 1943.
estimated it in 1942 without consideration of the age distribution
and again in the mid-1940s when
when age distribution is considered, in line with
, who estimated that the event likely occurred some
5 years later than indicated in the ice-core record.
The growth rate of atmospheric CO2 (dCATM/dt) corresponds to the
net balance between anthropogenic emissions and the ocean and terrestrial
fluxes:
dCATMdt=AGR=EFF+ELUC-O-L,
where EFF is the anthropogenic emissions of fossil-fuel
burning and cement production, ELUC the net CO2
emissions from changes in land use, and O and L the ocean and land sink
strength respectively. The total flux from the terrestrial biosphere is given
by
B=L-ELUC.
We define here the emission terms EFF and
ELUC as positive fluxes into the atmosphere and the sink
terms O, L, and B as positive fluxes out of the atmosphere.
Anthropogenic CO2 emissions
Fossil-fuel combustion and cement production
The Carbon Dioxide Information Analysis Center (CDIAC) provides annual
estimates of CO2 emissions from fossil-fuel burning, cement production,
and gas flaring (EFF) from 1751 to the present
. Here we use their most recent global estimates
between 1900 and 2000, which have an uncertainty of ±5 %. However, it
should be noted that an uncertainty range of about 11 % may be more realistic
when accounting for differences in the datasets and methods to estimate
CO2 emissions .
Emissions from LUC
The net CO2 emissions from changes in land use (ELUC)
are usually derived from information about changes in carbon stocks from
cropland cultivation or pasture expansion and abandonment, wood harvest,
shifting cultivation, deforestation/afforestation, and forest regrowth after
land abandonment. As this net flux cannot be directly measured, it is usually
estimated using models that track carbon stocks in the different pools from
inventories and historical accounts (the bookkeeping approach) or by
process-based models which simulate carbon fluxes due to imposed changes in
photosynthesis and decomposition processes. It is important to distinguish
between reconstructions of CO2 fluxes based on gross changes in land
use and those based on net changes, since the latter were found to
underestimate fluxes by more than 0.5 Pg C yr-1 .
Here, we use three different ELUC datasets, the first two
relying on gross land-use transitions:
The bookkeeping model of is used in the Global Carbon Budget assessment
and covers the period 1850–2005. It is mainly based on regional statistics from the Food and Agricultural Organization
(FAO, 2010) and includes the effect of
peat fires (from 1997 onwards) and fire suppression, the latter only for the
USA. The model by allocates pasture preferentially to
grassland, which may yield lower CO2 emissions by reducing deforestation
.
The “Bookkeeping of Land Use Emissions” (BLUE) model described by relies on the land-use
transitions from (which is based on the HYDE 3.1 database for cropland and pasture areas)
to reconstruct fluxes between 1501 and 2012 in a spatially explicit way. New cropland and new grassland are both taken proportionally
from natural vegetation types. Two subsets of ELUC are calculated, one using vegetation and soil carbon stocks from
and the other using the modifications proposed by , that feature generally lower carbon
densities for natural vegetation and lead to lower emissions. More details about the data sources and methods can be found in the original literature.
We also use LUC emissions estimated by a set of process-based models, described in Sect. 2.3.4. This is intended to account for
the loss of additional sink capacity, as discussed by .
Ocean and land sinks
Observation-based estimates of CO2 exchanges between the atmosphere, the
ocean, and terrestrial ecosystems are only available since the 1970s
. Here, we use
different reconstructions of the ocean and terrestrial sinks for the 20th
century, based on indirect methods. The goal of this procedure is two-fold:
to test the ability of these reconstructions to close the CO2 budget and
to gain insight into the drivers of the 1940s plateau.
Atmospheric CO2 growth rate (AGR) from the observational record,
calculated from the spline
fit in Fig. (black line, top panel). Fossil-fuel emissions from the CDIAC database and
respective uncertainty (EFF) and the reconstruction of ocean (OJ) and biospheric (BJ) fluxes from
(filled areas in top panel). The
resulting balance from the latter three datasets (AGRJ) and uncertainty is shown in the top
panel (dashed and dotted lines, respectively), and the corresponding difference
between AGR and AGRJ is shown in the bottom panel.
Double deconvolution of CO2 and δ13C records
used a double-deconvolution technique to reconstruct land
and ocean fluxes from measurements of atmospheric CO2 and δ13C
taken from the Law Dome ice-core record between 1800 and 1990. Their
analysis relied on a previous dataset of
the same CO2 ice-core record used here to solve two
mass-balance equations for atmospheric CO2 and δ13C, assuming
fixed ocean mixed-layer response. The method uses prescribed carbon
fossil-fuel emissions and their δ13C signature with a box model to
simulate isotopic disequilibrium fluxes between the atmosphere, ocean (OJ), and
biosphere (BJ), as shown in Fig. (top
panel). used the same measurements in a Kalman filter
double deconvolution. They came to the generally similar conclusion, namely
that the oceans played a significant role in creating the 1940s plateau.
These two double deconvolutions have some common weaknesses. Neither
calculation considers climate-driven variations in terrestrial isotopic
discrimination , which likely covary with
CO2 fluxes that are also driven by climate. The calculations also do not
consider changes in the distribution of C3 and C4 plants with time (Scholze
et al., 2008). Both of these effects may be important. It is possible to
calculate both effects with process models, generally as part of a forward
model calculation, but it would be problematic to calculate them in an
inversion such as a double deconvolution.
As with any inversion, the results depend on the choice of statistics such as
the magnitude of uncertainties and the degree of
smoothing of the fit to the ice-core measurements used in the mass balance
method . In both cases, these choices define how much of the
variability in the ice core is considered as “signal” to be interpreted and
how much is considered “noise” to be ignored. Such choices can be subjective
and lead to differences in the magnitudes of variations. The scatter in the
Law Dome δ13C ice-core measurements at the time of the plateau is
significant compared to the signal that we need to interpret to understand
the cause of the plateau. Furthermore, emissions from LUC have the same
isotopic signature as L, making it impossible to disentangle the two terms,
and fluxes from the C4 photosynthesis pathway (which has a lower affinity for
the lighter carbon isotopes) may be attributed to the ocean.
Nevertheless, double deconvolutions interpret measurements that represent
globally aggregated signals, allowing estimation of the main decadal
variability patterns in the land and ocean sinks due to changes in climate
over long timescales . These double
deconvolutions may thus be used to compare with the patterns found in our
model-based reconstructions. Here we use estimates of the ocean sink from the
double deconvolution by , i.e. OJ, as a
reference for natural variability in the ocean sink.
Reconstruction of anthropogenic CO2 uptake by the ocean
Several methods have been developed to estimate ocean CO2 fluxes from
observations
;
however, most of them cover only the last 3 decades of the 20th century.
used an inverse technique to reconstruct the oceanic
response to the anthropogenic perturbation, i.e. the uptake of anthropogenic
CO2 by the global oceans between 1765 and 2008. Their estimates of
oceanic CO2 uptake (henceforth OK) and their respective
uncertainties are shown in Fig. 3. In their reconstruction, the transport of
anthropogenic CO2 in the ocean is described by an impulse response
function, using a kernel that describes ocean circulation and allows us to
trace the transport of CO2 from the surface to the deep ocean. This
kernel is calculated from observations in recent decades of active and
passive tracers: temperature, salinity, oxygen, naturally occurring
14C, CFCs, and PO4. However, in their
approach ocean circulation does not include natural variability, apart from a
seasonal cycle. Nevertheless, their reconstruction is the one used in most of
the 20th century reconstructions and, despite not
representing interannual to decadal variability, it sets a reference level
about which we can define the range of ocean variability required to explain
the plateau.
Ocean sink from CMIP5 models
Currently, analysis of the role of ENSO in variations in oceanic sink
reconstructions from ocean general circulation models including
biogeochemistry and driven by climate and atmospheric CO2 observations
is only available for the second half of the 20th century
. One way of gaining insight into the
possible role of the ocean in explaining the plateau could come from the
analysis of coupled climate–carbon simulations over the 20th century. Despite
the fact that the simulated variability is not necessarily in phase with the
observed one, these simulations offer the opportunity to estimate the
potential amplitude and patterns of carbon flux variability at interannual to
decadal timescales.
We evaluate the ranges of natural variability in the global ocean sink using
outputs of global ocean CO2 flux from a set of 16 general
circulation models and and ESMs used for
the Coupled Model Inter-comparison Project Phase 5 (CMIP5), over the period
1860–2000. In order to match the timescales of the ice-core record, the
annual values of ocean fluxes were filtered according to the air-age
distribution for CO2 in DE08 ice and anomalies are
calculated as the departure from the 30-year moving average.
Some of the models differ only in their atmospheric resolution or the
representation of certain physical processes in the ocean, whose details are
given by . In the historical simulation, increasing
atmospheric CO2 concentrations were prescribed, as well as external
forcings such as sulfate aerosols, solar radiation variability, and volcanic
eruptions. When considering only one realization of each model, the internal
climate variability patterns and their influence in the resulting outputs may
not be fully captured . Therefore, we also evaluated global
ocean flux outputs from simulations using the same forcings as those
mentioned above but initialized with perturbed initial conditions. The
IPSL-CMA5 performed a set of six realizations of the historical simulation at
low resolution (IPSL-CMA5-LR), plus three realizations at mid-resolution
(IPSL-CMA5-MR), which may provide a better depiction of the ranges of natural
variability to be expected in the ocean sink. For these simulations we also
analyse variability in tropical sea-surface temperature, which allows
evaluation of the contribution of ENSO to the strongest anomalies in oceanic
uptake.
Summary of the dynamics global vegetation models used to estimate LDGVM. More details about the
way each model represents LUC may be found in .
Model
Spatial resolution
Vegetation
Fire
N-cycle
Reference
CLM4.5
1∘ × 1∘
Imposed
Y
Y
JULES
1.88∘ × 1.25∘
Dynamic
N
N
JSBACH
0.5∘ × 0.5∘
Imposed
N
N
LPJmL
0.5∘ × 0.5∘
Dynamic
Y
N
LPJ-GUESS
0.5∘ × 0.5∘
Dynamic
Y
N
OCN
1∘ × 1∘
Imposed
Y
Y
ORCHIDEE
2∘ × 2∘
Imposed
Y
N
VISIT
0.5∘ × 0.5∘
Imposed
N
Y
As in Figure but for the independent estimates of sources and sinks:
EFF from CDIAC, ELUC from Houghton (H), BLUE (B), and BLUE with
lower C-stock changes (Blc) and DGVMs forced with LUC, ocean from
reconstruction,
and land sink as estimated by DGVMs forced only by CO2 and climate. In the bottom panel,
the difference between observed AGR and AGRH, AGRB,
AGRBlc,
and AGRDGVM is shown.
Land sink from dynamic global vegetation models (DGVMs)
The land sink may be reconstructed with a DGVM forced with climate observations and atmospheric CO2 from
ice-core data, as performed in the TRENDY project , and used
in other reconstructions of the CO2 budget . These
models simulate water and carbon exchanges at the ecosystem level, and some
models also simulate vegetation dynamics, disturbance, and nutrient limitation
(Table ).
In experiment S2 from TRENDY, models are forced with climate observations
from the Climatic Research Unit and National Centers for Environmental
Prediction (CRU-NCEP v4) between 1900 and 2000 but do not represent LUC. Monthly net biome production fields from each model were integrated
globally and aggregated over each year to produce an annual time series for
the 20th century. Figure shows the annual values of the
global land sink as evaluated by the group of DGVMs (LDGVM).
Closing the CO2 budget
As discussed above, these different sets of data for the carbon budget terms
should, if correct, allow reconstruction of the Law Dome CO2 record
during the 20th century. The estimates of were originally
calculated from earlier measurements of the Law Dome record, which were
confirmed by new measurements. Therefore, atmospheric CO2 growth rate
calculated from fossil-fuel emissions and their ocean and biospheric fluxes
(AGRJ), i.e.
AGRJ=EFF-OJ-BJ,
should be similar to the AGR record resulting from the value obtained from
our spline fit on atmospheric CO2 concentration. However, it should be
noted that in the smoothing is stronger. The values of
OJ and BJ are shown in
Fig. together with the resulting AGRJ
between 1900 and 1990 and the corresponding difference with the observations
(i.e. AGRJ minus AGR, ΔAGRJ).
The other sets of data, being calculated using very different techniques, are
largely independent of the CO2 record and of each other. However, it
should be noted that in reality these fluxes are not entirely independent
of each other. For example, the emissions resulting from LUC will depend to
a certain extent on the carbon stocks of the terrestrial ecosystems, i.e. in
previous states of L. This is partially taken into account in DGVMs forced
with LUC, but not in the other datasets. The resulting CO2 budget using
the different datasets for each term may be calculated using
Eq. , EFF data and respective
range, the ocean uptake reconstruction from
(OK), the land sink from DGVMs (LDGVM), and
the four ELUC estimates, i.e.
AGRi=EFF+ELUC-i-OK-LDGVM;,
with i= (H; B; Blc; DGVM), referring to each of the four datasets used to
estimate emissions from LUC (Houghton, BLUE, BLUE with low
C-stocks, and DGVMs, respectively) and LDGVM refers to the inter-model median
of the global land sink from DGVMs (S2 experiment), shown in Table . To be compared with AGR from the ice core, the annual values
of AGR computed using Eq. () need to be smoothed in order to
match the air-age distribution of CO2 in air trapped in ice bubbles at
DE08 as proposed by .
Terrestrial sink during the plateau period (1940–1950) and during the El Niño event of 1940–1942, estimated
by the set of DGVMs. Values are in PgC yr-1.
Model
L (1940–1950)
L (1940–1942)
CLM4.5
0.79
0.81
JULES
0.72
-0.29
JSBACH
1.14
1.27
LPJ-GUESS
0.50
0.49
LPJmL
-0.43
0.09
OCN
0.89
0.30
ORCHIDEE
1.2
0.80
VISIT
0.85
0.57
It should be noted that the AGR in Eq. () suffers from
inconsistencies between terrestrial emission and sink terms when
ELUC is derived from bookkeeping rather than DGVMs:
while LDGVM includes the effects of changing environmental conditions, which
historically created a sink, the bookkeeping estimates assume that carbon
densities do not change over time, but keep them fixed at (the higher)
observational values from recent decades . This creates a
tendency towards overestimating early land-use emissions, likely some 10 %
for the industrial era . Furthermore, the bookkeeping
estimates do not include the loss of additional sink capacity
. DGVMs do include the loss of
additional sink capacity in their ELUC by using the S2
experiment of no LUC under transiently changing environmental
conditions as reference, so that the loss of the increased carbon stocks of
forests that are replaced by agriculture are attributed to
ELUC. While the effect of constant carbon densities in the
bookkeeping method leads to AGR being overestimated for earlier decades, the
effect of replaced sinks leads to AGR being underestimated. However, this
effect becomes significant only with the strong climate change after the
1950s .
The CO2 growth rate during the 20th century, calculated from each set of data, is shown in Fig.
with the corresponding departure from the observed values (ΔAGRi). We represent the uncertainty range
of the reconstructions as the uncertainty in EFF (±5 %), OK (reported by
),
and each individual ELUC estimate (±0.5 PgC uncertainty estimated by for
the bookkeeping models, and model spread for the DGVM).
The differences between each reconstruction and observations during the
period 1940–1950 are summarized in Table and provide an
estimate of a residual sink further required to explain the CO2
stabilization.
Difference between reconstructed and observed AGR (ΔAGR, in Pg C yr-1) during the periods 1940–1950
(positive values indicate an overestimation by the reconstructions). ΔAGRJ corresponds to the
reconstruction using EFF, OJ, and BJ as in Fig. and the
other ΔAGR values to the reconstructions based on EFF, OK, the inter-model median
value of the land sink estimated by DGVMs (S2), and the different estimations of ELUC, as in Fig. .
Set
ΔAGR
ΔAGRJ
0.1 ± 0.7
ΔAGRH
0.9 ± 0.8
ΔAGRDGVM
1.2 ± 1.0
ΔAGRB
2.0 ± 0.8
ΔAGRBlc
1.5 ± 0.8
Testing LUC with idealized experiments
The LUC component of the carbon budget is one of the most uncertain terms
and is as large as
EFF in the first part of the 20th century. The δ13C
record provides a constraint on the relative contribution of the oceanic and
terrestrial fluxes to the observed CO2 emissions. This allows evaluation
of the extent to which LUC processes could contribute to the
residual CO2 sink. Here, we perform a set of idealized experiments to
estimate the contribution of different terms of LUC to the overall carbon
balance, as well as their compatibility with the δ13C record.
We use an updated version of the relatively simple coupled carbon cycle and
climate model OSCAR to integrate the different components
of the carbon budget in a realistic mathematical and physical framework. The
version used is an alpha version of , but no significant
change to the bookkeeping module occurred during development. The model
includes a mixed-layer impulse response function representation of the ocean
carbon cycle . Carbonate oceanic chemistry is sensitive to
atmospheric CO2 and temperature change, and stratification is accounted
for by changing the mixed-layer depth according to sea-surface temperature
change, following CMIP5 models. The pre-industrial land carbon pools and
fluxes are calibrated on the multi-model average of the TRENDY v2 models
. Net primary production (NPP) then responds to varying
CO2 and climate, and heterotrophic respiration to varying climate, all
of which are calibrated on CMIP5 models. OSCAR embeds a bookkeeping module
capable of calculating its own CO2 emissions from
LUC, on the basis of land-cover change, wood harvest, and shifting
cultivation area inputs. LUC information is aggregated in 10
different regions from the original dataset from .
The variation in the stable carbon isotopic composition (δ13C) may
be calculated from the balance of the different CO2 fluxes
, which are simulated by OSCAR using
δ13Ct-δ13Ct-1=Ct-1×{δftEFFt-1+δot-1OAt+δlbt-1FBt-(δ13Ct-1+ϵo)AO-(δ13Ct-1+ϵlb)NPP}×Δt,
where t refers to time, C is the atmospheric CO2 concentration,
EFF denotes fossil-fuel emissions, OA and AO the gross
ocean–atmosphere and atmosphere–ocean fluxes respectively, FB
is the gross flux between terrestrial ecosystems and the atmosphere (i.e.
emissions from heterotrophic respiration, mortality, fires and land-use
change), and NPP is the global net primary production. In OSCAR all these
fluxes (in Pg C yr-1) are calculated as variations (e.g. ΔNPP)
from an initial state in 1700 (NPP0), whose values are given in Table , together with the values used for the fractionation
factors (ϵlb, ϵo) and isotopic composition of the
different reservoirs (δf, δo, δlb).
Constants and parameters used to calculate resulting δ13C from the OSCAR simulations.
Description
Value
Reference
NPP0, FB0
gross terrestrial fluxes in 1700
54 PgC yr-1
OA0, AO0
gross oceanic fluxes in 1700
73 PgC yr-1
δf
δ13C of fossil-fuel CO2
-24 (1750) to -28 (2010)
δo
δ13C of ocean surface water
2.5 (1750) to 1.5 (2010)
δlb
δ13C of the terrestrial biosphere
-25
ϵlb
isotopic fractionation of the terrestrial biosphere
-7
ϵo
isotopic fractionation between the air and ocean
0
The standard set-up of the OSCAR model does not capture the stall in
atmospheric CO2 during the 1940s, despite performing relatively well
during most of the 20th century. This failure may be due to a variety of
reasons, as discussed by . Nevertheless, it allows us to
track the individual contribution of each budget term to the overall CO2
budget. Here, the OSCAR model is used to evaluate the relative effect of
hypothetical extreme LUCs during 1940–1950 in AGR, for instance
related to the abrupt socioeconomic changes imposed by the Second World War (and prolonged
during the early post-war period) in many regions. Our idealized experiments
exaggerate the magnitude of the hypothetical LUC during 1940–1950, but they
allow quantification of their relative impact on the resulting AGR and
δ13C, as compared with the standard OSCAR set-up, providing an
indication of how much a given LUC transition may contribute to the global
carbon balance during the period 1940–1950.
The global area under LUC transitions during 1940–1950 used in the default
OSCAR set-up is shown in Table . In the first test (T1), we
set all the transitions from forest to other land-cover types to 0 between
1940 and 1950, i.e. artificially and abruptly stopping deforestation over
the globe in 1940. The second test (T2) doubles the area corresponding to
forest expansion in each year (T2). The third test prescribes a halt in all
expansion of cropland and pasture areas (T3), which indirectly also sets
deforestation to 0, since forest is only lost to either crop or pasture
(Table 5); the last test is to stop all wood harvest (T4).
Results
Reconstructions of CO2 sources and sinks
The record of emissions from fossil fuel and cement production during the
20th century (Fig. ) shows a slow increase of
EFF at a rate of ca. 0.02 Pg C yr-2 during the first
4 decades, punctuated with periods of slight decrease. From 1940 to 1950,
EFF was on average 1.4 Pg C yr-1 and, in spite of a
small decrease during 1945–1946, even accelerated, with a rate of change of
0.05 PgC yr-2 during the full period. As uncertainty in
EFF is also very small (<±0.1 Pg C yr-1) in the
first half of the 20th century, and a stabilization of
CO2 would imply EFF being 0 for the whole period, its
role in explaining the CO2 stabilization in the 1940s is excluded. Here
we evaluate whether the available sources of data about the other terms of
Eq. () allow reconstruction of the plateau.
Average global LUC transitions (in Mha yr-1) during 1940–1950 from , used in the OSCAR model default set-up.
Transition to
Transition from
Desert and urban
Forest
Grassland and shrubland
Cropland
Pasture
Desert and urban
–
–
–
1.6
9.5
Forest
–
–
–
2.1
4.9
Grassland and shrubland
–
–
–
4.4
15.0
Cropland
0.2
0.7
0.8
–
2.4
Pasture
2.6
6.2
4.6
1.5
–
Given that the estimates of ocean and biospheric fluxes from ,
OJ and BJ, were calculated using a previous
version of the CO2 record used here, they are expected to reproduce the
observed variations in CO2, as given by the general agreement between
observations (AGR) and reconstructed (AGRJ) shown in
Fig. (top panel). However, in spite of the uncertainty
limits of AGRJ encompassing the observations, discrepancies
are found for some periods of the century, as the one from 1930 to 1940, and the
one from the late 1960s until 1980. These are more evident when analysing the
difference between reconstruction based on and observations,
ΔAGRJ (Fig. , bottom panel). This is
likely due to the different degrees of smoothing used in and
here, as exemplified in Fig. S1.
The reconstructions performed using the different ELUC
estimates, the ocean sink from , OK, and
the terrestrial uptake from DGVMs (LDGVM) are shown in
Fig. (top panel). The discrepancies between
reconstructions and observation (Fig. , bottom panel)
present variability patterns with different timescales, with a deviation
from 0 beginning around 1910, increasing up to a maximum ca. 1950, then
decreasing back to 0 around 1990 and decadal variability superimposed on this longer-term variation. All reconstructions overestimate AGR between 1940 and
the mid-1970s, and this overestimation is particularly large around 1945 and
1960. Results from ELUC-H and ELUC-DGVM are
similar during most of the century and lead, generally, to lower
discrepancies between reconstructions and observations, as compared with
ELUC-B data. In the case of the two BLUE datasets, AGR is
overestimated during most of the 20th century and by up to
1–2 Pg C yr-1 in two periods: the 1940s and 1950s–1960s.
During the plateau period (Table ), the values of
ΔAGR are, as expected, lower for AGRJ than for the
other datasets, although the absolute uncertainty of the reconstruction is
1 order of magnitude higher (1.4 Pg C yr-1) than the estimated misfit
(ΔAGR, 0.1 Pg C yr-1 ), i.e. the extra sink required to match
observed AGR. In AGRJ, the uncertainty is likely
overestimated because OJ and
BJ have anticorrelated errors in the double deconvolution. If this had been taken into
account, uncertainty would be similar to the one in their CO2
observations, while it is generally of the same magnitude as the estimated
values, still increasing from 1960 onwards.
For the independent datasets, the mismatch with observations is smaller for
ELUC-H and ELUC-DGVM (0.9 and
1.2 Pg C yr-1, respectively), reaching 2.0 Pg C yr-1 for
ELUC-B. Part of the discrepancy observed during the plateau
period in AGR estimated using Houghton and BLUE datasets results from the
consistently higher values of BLUE over the whole century. The relative
variation in ΔAGRB and ΔAGRBlc
during 1940–1950 relative to the 1920s and 30s roughly matches the one
observed in the other two datasets.
However, it should be noted that DGVMs also differ considerably in their
estimates of the land sink during 1940–1950, with one model (LPJmL) even
estimating a terrestrial source rather than a sink during the period (Table ).
Testing the hypothesis for the plateau
As shown previously, the datasets of carbon budget terms
(EFF, OK, LDGVM, and the
different ELUC data) lead to overestimation of AGR by 0.9 to
2.0 Pg C yr-1 for ELUC-H and ELUC-B
respectively. Thus, there is a sink missing in the budget that could be
explained by (i) decadal variability in the ocean sink not represented in
OK; (ii) processes absent from all the TRENDY models causing
extra land uptake in ecosystems without LUC; or (iii) LUC
processes that lead to carbon uptake and are not, or not sufficiently,
included in the current datasets.
Ocean variability
Despite both ocean reconstructions (OJ and
OK) generally agreeing on the long-term trend of the ocean
sink (Figs. and ), OK
presents a smooth increase, consistent with the evolution of atmospheric
CO2, while OJ points to the existence of large
multi-decadal variations superimposed on the increasing trend, the largest of
them coinciding with the plateau period. During 1940–1950 the two datasets
differ by about 0.5 Pg C yr-1 (OK ca. 0.7 Pg C yr-1
and OJ ca. 1.2 Pg C yr-1 on average), providing a
reference value for the possible contribution of natural variability in the
ocean to the sink required to stabilize atmospheric CO2.
It is important to evaluate whether an enhancement of the ocean sink of the
magnitude reported by is likely to have occurred during the
first half of the 20th century. Such reinforcement of oceanic CO2 uptake
could only be explained by natural variability, as given by the large
difference between OJ and OK for this period
(0.5 Pg C yr-1), for instance due to a strong El Niño event, as
suggested by previous works for the
exceptional 1940–1942 El Niño .
This may be tested by evaluating the variability patterns of CO2 fluxes
in the global ocean calculated by the set of 16 ESMs from CMIP5 for the
historical period. Although the models are not expected to reproduce the
exact temporal evolution of the ocean sink because they simulate their own
climate variability, it is possible to test their ability to represent
decadal departures of magnitude of 0.5 Pg C yr-1 from the long-term
trend (as the difference between OK and OJ)
or up to 2.0 Pg C yr-1 (if we consider the residual sink to be in the
ocean).
Variability in the global CO2 uptake by the oceans, estimated by the group of CMIP5
climate models for the historical simulation, with prescribed atmospheric CO2, as well as solar
radiation variability, sulfate aerosols, and volcanic eruptions. The annual values of the ocean fluxes
are filtered using the same smoothing as the one applied to AGR, based on the air-age distribution for CO2 at
DE08 from .
The anomalies of the ocean sink calculated by the models for the historical
simulation (prescribed atmospheric CO2 and external forcings), filtered
to match the ice-core air-age distribution are shown in Fig. . As data are smoothed, the anomalies correspond to a
long-term pattern rather than an annual anomaly. The variation ranges
estimated by the models are about half of those suggested by
OJ, with most anomalies being smaller than
±0.15 Pg C yr-1, although in some models (e.g. GISS-E2-R-CC and
IPSL-CM5A-MR) anomalies may reach values of about ±0.2 Pg C yr-1.
The anomalies in ocean CO2 uptake present multi-decadal variations which
are consistent among the 16 models and are due to the ocean response to the
CO2 forcing. In particular, during the plateau period, most models
estimate lower ocean uptake because of the slow-down of the anthropogenic
perturbation. The inter-model comparison indicates that, assuming the
magnitudes of variability of the modelled ocean fluxes are representative of
the real ocean, an anomaly of more than ca. 0.2 Pg C yr-1 in the ocean
sink is unlikely to be registered by the ice-core record.
Maximum decadal anomalies of ocean CO2 uptake in the IPSL-CMA5 simulations (PgC yr-1 per decade)
and corresponding anomaly in tropical SST (∘C) in the Nino3.4 region. The annual values of the ocean fluxes are
filtered using the same smoothing as the one applied to AGR, based on the air-age distribution filter from .
The SST anomaly is calculated as the average departure of the filtered SST data from a 30-year-long reference period.
Realization
Time
Oanom
SSTanom
IPSL-CMA5-LR r1i1p1
1932
0.08
0.01
IPSL-CMA5-LR r2i1p1
1885
0.06
-0.06
IPSL-CMA5-LR r3i1p1
1889
0.11
-0.08
IPSL-CMA5-LR r4i1p1
1930
0.13
0.01
IPSL-CMA5-LR r5i1p1
1991
0.14
-0.06
IPSL-CMA5-LR r6i1p1
1895
0.12
0.02
IPSL-CMA5-MR r1i1p1
1991
0.22
-0.05
IPSL-CMA5-MR r2i1p1
1918
0.20
-0.06
IPSL-CMA5-MR r3i1p1
1976
0.17
-0.02
As in Figure but for six different realizations from IPSL-CM5A-LR and three from
IPSL-CM5A-MR (top panel) and the corresponding SST temperature anomalies in the Nino3.4 region. The SST anomaly
is calculated as the 10-year moving average departure of the SST data from a 30-year-long reference period.
Nevertheless, to account for the impact of natural variability in ocean
fluxes it is advisable to consider a larger number of realizations for each
model, given that results may differ considerably, especially in the
timescales of interest to this study . The global ocean
CO2 uptake estimated by six realizations from IPSL-CMA5-LR and the three
from IPSL-CM5-MR is shown in Fig. a. Some of the different
simulations reveal strong decadal variations, with anomalies varying (in some
cases) by ±0.3 Pg C yr-1. These variations are more pronounced for
the model with higher spatial resolution (IPSL-CMA5-MR), suggesting a
possible influence of smaller-scale processes that control internal
variability of the ocean, for instance better representation of the
westerlies in the Southern Ocean . Nevertheless, such a
range of variation is consistent with observation-based estimates for the
late 20th century .
The strongest positive anomalies in the ocean sink for each of the IPSL-CMA5
simulations and the corresponding peak year are presented in
Table , together with the corresponding variations in the
Tropical Eastern Pacific sea-surface temperature (SST; Fig. b and
Table ). Only three out of the nine simulations present
strong ocean uptake coincident with warming (but very feeble) of the tropical
oceans: r1,r4 and r6 from IPSL-CMA5-LR. This is consistent
with the reduced upward transport of carbon-rich water from the deep ocean,
associated with weaker upwelling due to the persistence of warmer surface
temperatures (as during El Niño events). However, it is not possible to
establish a straightforward link between tropical Pacific SST and the
enhancement of the ocean sink for any of the other simulations.
Land response to climate
If the additional sink were provided by land, and considering the inter-model
median LDGVM of 0.8 Pg C yr-1, a total terrestrial
uptake of more than 1.5 Pg C yr-1 would be needed. This magnitude is
comparable to the average land sink in the early 2000s (1.3 Pg C yr-1)
estimated by atmospheric inversions , when the effects of
CO2 fertilization are already important .
It is thus worth testing whether DGVMs capture realistically the response of
terrestrial ecosystems to the climate forcing during the plateau period, as
well as to the strong El Niño event (1940–1942), as ENSO impacts on regional
climate and terrestrial ecosystems have been studied for later events
and therefore provide a known reference to
analyse the expected anomalies in the climate forcing and the corresponding
simulated response.
Table shows the average land sink estimated by DGVMs during
1940–1950 and the 1940–1942 El Niño. DGVMs estimate in general a relatively
strong terrestrial sink during the plateau, except LPJmL, which simulates a
0.43 Pg C yr-1 terrestrial source during the period. When compared to
the period 1900–1930, all DGVMs estimate an increased sink in the Northern
Hemisphere, especially at high latitudes, coinciding with generally warmer
and wetter conditions throughout most of North America and Eurasia
(Fig. ). This increased sink is mostly due to
strong enhancement in gross primary productivity (Fig. S3), consistent with
the increased growth observed in tree rings in the Northern Hemisphere
. In the tropics, models diverge significantly in the
anomalies in CO2 uptake in response to the temperature (generally lower)
and precipitation (above average in most regions) patterns during the
plateau. Differences in model sensitivity to temperature and precipitation,
or lack of proper fire representation, may explain part of this mismatch.
Five of the nine models indicate a reduction of terrestrial uptake in 1940–1942
(as compared to the plateau period), expected during a warm ENSO event
although not as strong as the response of the terrestrial sink to El Niño
registered in the late 2000s .
In general, temperature anomalies (Fig. S4, left panel) over land in 1940–1942 present an El Niño-like distribution
, with warming in most of the tropical and subtropical regions, and the strong cooling
over Europe reported by . However, dry conditions during 1940–1942 in the forcing are confined to
part of northern South America and the Philippines, rather than the characteristic overall drying of part of Amazonia
and subtropical South America, southern Africa, or Australia that usually leads to weaker CO2 uptake by land
ecosystems during positive ENSO events .
Response of the terrestrial ecosystems to the climate anomalies during the plateau period, simulated by the
DGVMs. Temperature (left top) and precipitation (left bottom) anomaly fields during 1940–1950 (relative to 1900–1930)
and the corresponding latitudinal anomaly of LDGVM estimated by each model (grey lines) and the
multi-model average (right panel).
Although most models capture the reduction in terrestrial uptake in the
tropical regions (Fig. ), some estimates of tropical
anomalies are very small. At the same time, most models estimate a strong
enhancement of the sink in northern latitudes, especially in the Eurasian
region, which partially offsets the small decrease of CO2 uptake in the
tropics. The enhanced northern CO2 uptake during El Niño derives from
a combination of high photosynthesis in North America (where strong warming
is registered) and a combination of enhanced photosynthesis and low
respiration in Europe (which registers negative temperature anomalies during
all seasons except summer). The very strong response to the latter effect in
some models explains the very small land sink anomalies found for most of the
models and the enhanced sink identified by CLM4.5, JSBACH, and LPJmL.
Average difference in ELUC and atmospheric CO2 growth rate between the OSCAR standard run and
the simulations using different LUC hypothetical scenarios during 1940–1950, in Pg C yr-1.
Test (from 1940 until 1950)
Δ ELUC
Δ AGR
T1 – stop deforestation
0.51
0.39
T2 – double forest expansion
0.07
0.06
T3 – stop crop and pasture expansion
0.49
0.37
T4 – stop wood harvest
0.23
0.17
Land-use change
The differences in ELUC for the four extreme hypothetical
scenarios and the standard OSCAR run are shown in Fig.
(top), as well as the comparison of the resulting changes in the atmospheric
CO2 growth rate (centre) and δ13C (bottom) with the
observational values. The average differences in LUC emissions and resulting
AGR during 1940–1950 are summarized in Table .
The two largest reductions in ELUC result from halting either
deforestation or crop and pasture expansion, which lead to an average
reduction of 0.51 Pg C yr-1 and 0.49 Pg C yr-1 during the decade,
peaking at about 0.8 in 1950, when the standard OSCAR LUCs are resumed. The
loss of forest to cropland and pasture influences the fluxes resulting from
crop and pasture expansion, as shown by the small differences between the two
emission trajectories (T1 and T3). Due to the interactions between these two
transitions and the land sink, the resulting difference in atmospheric
CO2 is about 25 % smaller: 0.38 and 0.36 Pg C yr-1 for T1 and
T3 respectively.
Resulting ELUC from OSCAR simulations for hypothetical scenarios about changes in LUC
during 1940–1950 (top). In T1 forest conversion is set to 0 (green solid), in T2 the rate of forest expansion
during the period is doubled (green dashed), in T3 cropland and pasture expansion are stopped (yellow solid), and
in T4 wood harvest is set to 0 (yellow dashed). The ELUC from each test is compared with the
LUC emissions in the standard OSCAR simulation (red). The atmospheric CO2 growth rate (AGR) resulting from
standard OSCAR and each test is compared with the ice-core record (centre). The δ13C values
corresponding to each test (bottom) are compared with δ13C from the ice-core record and the
corresponding uncertainty (markers and error bars).
Stopping wood harvest during 1940–1950 (T4) leads to ELUC
0.23 Pg C yr-1 lower than the standard simulation, resulting in AGR
differences of 0.17 Pg C yr-1. However, in this case,
ELUC increase rapidly from about 1950 onwards and even
surpass the values estimated by the standard simulation, which may be related
to the predominance of biomass burning and fast decomposition processes
during the first years after resuming harvest. Although smaller in magnitude,
a hypothetical doubling in the area under forest expansion (T2) leads to a
decrease in ELUC of 0.07 Pg C yr-1, impacting AGR by
0.06 Pg C yr-1.
The relative abundance of carbon isotopes 12C and 13C depends on
the carbon fluxes between the different reservoirs, as the driving processes
(e.g. photosynthesis, fires, respiration, ocean dissolution) have specific
isotopic fractionation ratios . The isotopic signature of
carbon in CO2 samples (usually expressed as δ13C) thus
provides a constraint on the relative contribution of each process to the
observed variations in atmospheric CO2 concentration. Isotopic data from
the ice-core record reveal a flattening of δ13C between ca. 1915
and 1950 .
The δ13 calculated using the standard OSCAR set-up generally
remains within the uncertainty range of the observations, except during
1950–1960 and the late 1990s. In spite of performing rather well for most of
the century, the standard set-up does not fully capture the flattening of the
δ13 record during the 1915–1950 period.
An increase in δ13 during the plateau period is observed for all
the idealized experiments, consistent with an increased terrestrial sink.
Despite our tests imposing changes in ELUC only for the
1940–1950 period, differences between their δ13C signature and the
one from the standard set-up are noticeable until the late 1960s. Experiments
T1 and T3 lead to a stronger increase in δ13C relative to the
standard simulation but still remain roughly within the uncertainty limits
of the observations between 1940 and 1950 and actually remain closer to
observed δ13C in the subsequent decade.
Discussion
We find that the datasets of anthropogenic CO2 emissions combined with
the reconstructions of carbon uptake by terrestrial ecosystems and the ocean
are not able to reproduce the decrease in atmospheric CO2 growth rate
between 1940 and 1950 registered in the observations. A further sink of at least
0.9 Pg C yr-1 is still required. While uncertainty in emissions from
fossil fuels is much smaller than the sink required, uncertainty in the other
terms is very high.
CO2 sinks during the plateau
An ocean sink of about 1.2 Pg C yr-1 during the 1940s, as in the
dataset, is needed to explain partly the observed CO2
plateau; such a sink is compatible with the occurrence of a strong ocean
uptake anomaly due to natural climate variability superimposed on the
anthropogenic perturbation trend. The variation range of different
realizations of the IPSL-CMA5 model forced with perturbed initial conditions
is within the variability range found for ENSO impacts on oceanic CO2
uptake in the late 20th century (0.1–0.5 Pg C yr-1;
). Other works have suggested an ocean
uptake of 2–2.5 Pg C yr-1 during the 1940s or in
response to later ENSO events , which appears to be too
high in light of the variations simulated by the models and the recent
estimates from atmospheric inversions.
The role of an extreme ocean uptake event as, for instance, in response to
the 1940–1942 El Niño, does not seem likely to have been the sole driver of
the plateau – other sources of variability from the ocean may need to be
considered. have analysed unforced natural variability
in the ocean using century-long simulations from a set of six ESMs (a subset
of the ones we use in this study). At interannual to decadal timescales,
models indicate a strong contribution of the Southern Ocean to the global
ocean sink due to (1) variations in wind stress and deep-water upwelling
controlled by the Southern Annular Mode (SAM) and (2) the occurrence of deep
convective events that trigger a reduction in sea-ice coverage and intense
mixing of surface waters with carbon-rich deep waters. Regarding SAM-induced
variability, the changes in atmospheric circulation in the late 2000s have
been recently linked to a remarkable increase in CO2 uptake by the
Southern Ocean, from about 0.6 Pg C yr-1 in 2002 to 1.2 Pg C yr-1
in 2011 . If variations of this order of magnitude
in the Southern Ocean would be accompanied by non-cancelling anomalies in the
tropical Pacific, one could expect a higher contribution of the ocean to the
global sink during the 1940s. Regarding the convective events, climate models
suggest a long multi-decadal timescale, from 20–30 to 50–60 years depending
on the model, which makes them relatively rare events even for a century-long
record. Satellite observations indicate the existence of such a deep
convective event in the 1970s , but there are no
observations for the 1940s. Given the lack of observation-driven datasets
able to capture these variabilities, this hypothesis remains speculative.
Considering a contribution of 0.5 Pg C yr-1 due to natural variability
in the ocean, as estimated by and recent observations, a
further (terrestrial) sink of 0.4–1.5 Pg C yr-1 is required. DGVMs used
to characterize the land sink during the 20th century indicate that
terrestrial ecosystems constituted an important CO2 sink, taking up
about 0.8 Pg C yr-1 during the period between 1940 and 1950 in response
to generally warmer and wetter conditions. The models estimate a small
decrease of the terrestrial sink during the strong El Niño event of 1940–1942
(and even enhancement in some models), which is not fully consistent with the
more recent observations of the terrestrial response to ENSO
. Despite most models capturing a decrease
in CO2 uptake in the tropics during the El Niño in response to
dryness, the reduction is likely underestimated, as DGVMs are known to have
problems in representing fire disturbance . In
any case, the aforementioned discrepancies would further reduce the
terrestrial sink rather than helping to explain the enhanced sink needed. In contrast, the inconsistency of the land-sink response to El Niño with
recent observations may also be due to the climate forcing, which does not
represent the characteristic drying pattern over most of the Southern
Hemisphere.
Before 1950, especially during the Second World War, the global meteorological network
coverage was poor in comparison with the late 20th century. Moreover, DGVM
simulations rely on CRU-NCEP v4, which uses the CRU dataset for monthly data
and NCEP/DOEII reanalysis to generate 6 h variability. As NCEP does not
extend to earlier than 1948, to generate the 6 h variations in CRU-NCEP
v4, the variability of a random year between 1948 and 1960 is applied to each
year before 1948. This may partly explain why the quality of the simulations
before 1948 is not as good as afterwards.
If precipitation was higher than average during this El Niño event in the
regions that usually experience drought instead (as indicated in the CRU/NCEP
data), the reasons for these opposite anomalies could be understood.
, using a 700-year reconstruction of Nino3.4, have shown that the
later decades of the 20th century were characterized by unusually high ENSO
variability, while the 1940s registered a peak of low ENSO variance. During
this period, have found a break in the correlation of
precipitation in south-eastern Australia and ENSO, associated with a positive
phase of the Interdecadal Pacific Oscillation . The
modulation of ENSO teleconnections in remote areas may imply a variable
relationship between ENSO and the terrestrial sink that deserves deeper
attention. Finally, it should also be noted that the 1940–1942 very strong El
Niño was followed by more than a decade with predominant La Niña
conditions , coinciding with a negative phase of the
Pacific Decadal Oscillation , which may explain the
persistence of an increased terrestrial sink during the plateau period.
The contribution of LUC
The different estimates of emissions from LUC differ
significantly during most of the 20th century, but their estimates
diverge to a greater extent in the earlier decades of the 20th century. We
find that the LUC emission estimates from the latest inter-model comparison
exercise (TRENDY v4) present good agreement with the bookkeeping data from
. The two BLUE datasets differ with
the former two datasets by up to 1 Pg C yr-1. The discrepancies between
BLUE and the other datasets likely result from the use of different
methodologies, definitions, and assumptions in each study , such as the definitions of pasture areas or the
way gross transitions are estimated. Moreover, the closer agreement of
ELUC-H and ELUC-DGVM is incidental, as the
models differ in the processes represented and definitions used
. Such differences are
considerable and their impact is of similar magnitude as, for instance,
stopping deforestation or wood harvest completely during 1940–1950, as
estimated in idealized simulations using the OSCAR model. It is notable that
the model estimates based on HYDE 3.1 show a stagnation of
the previously rising land-use emission rates during the 1940s. The BLUE
model shows that, globally, emissions from cropland and pasture expansion slow
down during the 1940s, while CO2 uptake in abandoned land increases
steeply. Net carbon sinks due to LUC are thus created in parts of
Europe, North America, and China, but they are not large enough to create an
overall sink in the terrestrial biosphere (Fig. ).
Land-use reconstructions rely on national inventories and agricultural
statistics. While these sources of data are expected to have become reliable
in recent decades, even then contradictory statistics are found at the
country level and between reported and satellite-based estimates
. For the early 20th century statistics of deforestation,
land abandonment, and agricultural area are expected to be highly unreliable
in many regions due to the lack of inventories, e.g. in Amazonia
. has shown that revisions in recent
inventories could account for regional differences in ELUC of
about 0.3 Pg C yr-1.
A major uncertainty results from all model studies applying land-use
reconstructions that are based on FAO data for agricultural areas, which are
available only from 1961 onwards. While included
additional historical sources for some regions, the HYDE 3.1 database
and thus the dataset by rely on
extrapolating these country-level statistics back in time using population
dynamics. In HYDE 3.1, the cropland and pasture values per capita are allowed
to change “slightly” prior to 1961 . Although
changes in per-capita values between the 1940s and 1960s amount to only
1 ‰ when averaged over all countries, they may be as high as
50 % in individual countries.
The uncertainty in the LUC emissions during the Second World War period thus remains high.
Although statistics about food production, population, or industrial output
were kept because of their direct interest to the war effort management
, information about other processes relevant for
ELUC may not be accurate (e.g. the impact of population
mobilization for war and industry on land abandonment, changes in wood
harvest).
For example, statistics for agricultural areas in the Soviet Union during
1940–1945 are almost absent. report a 6.6 Mha decrease of
crop area between 1940 and 1950 in the former Soviet Union, but
estimated a reduction in crop area in the territory of the
Russian Federation of 27 % or about 25 Mha for the same period. The number
of war-related deaths is estimated to be 26.6 million people, about 14 % of
the population , and agricultural output is estimated to
have fallen by up to 60 % during the peak of the war .
Furthermore, with the re-location of the industry from the western front to
the eastern provinces, about 10 million people are estimated to have been
evacuated from the western areas . Thus, the abandonment of
cropland might be even higher for the most affected war territories of
Ukraine and Belarus, where agricultural production was severely reduced
due to a shortage of manpower and destruction of infrastructure. The
interruption in agricultural production extended beyond the war period,
recovering only slowly. The crop area in Russia returned to the pre-war level
only in the early 1950s . Also, at the time of the Second World War, the
reliance of Russian population on fuel wood was likely much larger than in
the last decades of the Soviet Union. In China, the war-related mortality
during the Second World War in China is estimated to be of about 14 million people
and mass migration movements were also reported
. The cropland area likely decreased during the war period
and only started to recover after 1949, according to Chinese Historical
Cropland Database, which is not represented in HYDE 3.1 dataset
. A decade of reduced agricultural production and harvest in
the war-stricken regions, not accounted for in the HYDE 3.1 dataset, could lead
to substantial missing carbon uptake during this period.
The analysis of δ13 signatures corresponding to each of the
idealized experiments shows that differences in LUC datasets as
extreme as the ones tested here could still be compatible with the observed
δ13 record, although changes in land use of the magnitude of our
idealized tests are unlikely. Furthermore, effects of agricultural
abandonment and halting of deforestation due to historical events have little
effect on atmospheric CO2 when persisting only for short periods of time
(few decades or less), as model experiments suggest delayed emissions from
past LUC, in particular from soils, persist, and regrowth takes
time to reach its full potential .
concluded that the stalling of atmospheric CO2 during the 1940s was
unlikely to have been caused by LUCs. Still, the sensitivity
experiments with OSCAR suggest that it is reasonable to expect that events
not well represented (or included at all) in the current LUC reconstructions
may provide a non-negligible fraction of the 0.4–1.5 Pg C yr-1 required
for reconstructions to match the CO2 record during the period.
Other sources of uncertainty
Another process that could potentially contribute to a further increase in
the terrestrial sink is the impact of nitrogen deposition in net primary
productivity. have shown that nitrogen deposition
stimulated carbon sequestration in temperate forests in the USA during the
1980s and 1990s, with stronger sensitivity of carbon accumulation to lower
levels of nitrogen inputs. The authors estimated that nitrogen deposition
could increase carbon storage in ecosystems by ca. 0.3 Pg C yr-1.
The increase in fossil-fuel burning due to industrial expansion and the
beginning of the automobile era produced strong changes in nitrogen
deposition. The strong initial response of plants to the nutrient input could
have produced a sudden increase in the terrestrial sink, followed by
saturation due to soil acidification as deposition rates persisted
and other limitations such as phosphorus came into play
. At present, DGVMs still struggle to represent
realistically the interactions between ecosystems and between the nitrogen and
phosphorus cycles. Nevertheless, the DGVMs used here that include the
nitrogen cycle (CLM4.5, OCN and VISIT) estimate very similar values for the
terrestrial sink during the plateau period, and slightly stronger than the
inter-model mean (Table ).
Conclusions
This work has used the currently available estimates of sources and sinks of
CO2 during the 20th century and their associated uncertainties to gain
insight into the temporary stabilization of atmospheric CO2
concentration observed during the 1940s until mid-1950s, as well as
evaluating the mechanism previously identified as the main driver of such
stabilization.
Our results show that, although the oceans are likely to have contributed,
they cannot by themselves provide the complete explanation of the 1940s
plateau. A strong terrestrial sink is also required to match the observed
stalling in atmospheric CO2 during the period. Further work is required
to narrow the uncertainty in the carbon budget components in order to
identify other processes that might help to explain the 1940s plateau.
However, the discrepancies between observations and the carbon budget
estimated using independent reconstructions of each component are not
particular to the 1940s. This indicates that efforts to narrow down the
uncertainty of each term of the carbon budget are required.
The relationship between reconstructed terrestrial and ocean fluxes with the
climate anomalies observed during the early 20th century deserve greater
attention. Given the large difference between estimates of ocean flux
anomalies in response to climate variability, a new initiative is needed to
better characterize CO2 fluxes in the ocean during the 20th century,
e.g. by forcing the ocean circulation models with climate reconstructions.
In the case of the terrestrial sink, other processes currently not included
in the models or in the LUC reconstructions may have contributed to the
plateau. The effects of fire occurrence, changes in nutrient availability and
the devastating socioeconomic consequences of the Second World War are examples of processes
currently not well represented in the models.
It should be noted that the high-resolution Law Dome record is unique in its
precision and quality. However, the large measurements errors in even the
best δ13C ice-core data currently available make it difficult to
accurately quantify variations in the oceanic and terrestrial sinks. In high
accumulation sites such as DE08, new measurements of δ13 with
improved accuracy should reveal the high-resolution information contained in
the ice sheet and reduce the scatter of current estimates. It would also be
advantageous to get another insight into atmospheric CO2 and
δ13C changes during 1940s from a second high-resolution core.
This study thus allows us to identify a number of key aspects of the global
carbon budget that require deeper attention, if we are to better characterize
the coupled carbon–climate variability in the 20th century.