Introduction
The amplitude of the mean annual cycle of atmospheric CO2, an indicator of
seasonal terrestrial and ocean carbon exchange, has increased
over the Northern Hemisphere (NH) since observational records began in the
late 1950s . The
largest increases of 40–50 % were observed over the northern high
latitudes via surface monitoring and from
aircraft observations of the free troposphere .
The amplification of the atmospheric CO2 annual cycle primarily reflects
enhanced net exchange of CO2 with land surfaces rather than with the ocean
. Land–atmosphere CO2 exchange is highly
seasonal, especially in the NH mid- and high latitudes where photosynthesis
draws down CO2 in the spring and summer, and net ecosystem respiration
returns CO2 to the atmosphere e.g.,.
Because atmospheric CO2 observations are characterized by high precision
and accuracy, the gradual, multi-decadal increase in the seasonal amplitude
provides a unique observational target for Earth system models (ESMs)
intended to predict the long-term coevolution of climate and the carbon cycle. ESMs enable
the study of long-term effects of natural and anthropogenic forcing on the
terrestrial carbon cycle since they include mechanistic representations of
the carbon cycle by coupling land surface models that explicitly resolve
biogeochemical processes with models of the atmosphere, ocean, and other
components of the climate system . An advantage of
using a coupled model is that feedbacks between the physical climate and
biogeochemistry are represented in a self-consistent framework. This is
crucial since carbon fluxes are inherently linked to the physical climate;
for example, a change in gross primary productivity (GPP) will be associated
with changes in evapotranspiration, which feeds back on metrics such as
humidity, cloud cover, and precipitation. Moreover, in a fully prognostic
model, both climate and carbon cycle diagnostics are free to evolve rather
than being tied to input data sets that reflect the contemporary climate. The
mechanisms embedded in ESMs to predict future carbon–climate interactions
have been identified as the likely drivers of the observed mean annual cycle
amplitude increase as described in the following paragraphs.
The magnitude of the amplitude increase suggests a dominant role for enhanced
primary productivity during the growing season in addition to increased
CO2 release during the dormant season .
Greater atmospheric CO2 may facilitate plant carbon uptake through
increased water use efficiency , and results from
and suggest that
CO2 fertilization contributes at least 10 % to the CO2 mean annual cycle
amplitude trend. Human activity has not only increased the atmospheric
concentration of atmospheric CO2, but also modified reactive nitrogen
deposition (N deposition) in ecosystems. Deposition of nitrogen
oxides (NOx) and ammonia from combustion, livestock, agriculture, and
industrial sources may augment the supply of soil nitrogen available for
fixation by plants , alleviating a limitation on
terrestrial GPP .
Climate-change-induced warming and lengthening of the growing season may also
stimulate GPP and increase the seasonality of net exchange.
proposed that increased terrestrial CO2
uptake from a longer high-latitude growing season has driven the
amplification of the CO2 annual cycle since the trends in the CO2
annual cycle amplitude strengthen moving northward, and the greatest warming
has occurred during the winter and spring over the northern high latitudes.
Findings by ,
, and
support the hypothesis that longer growing seasons enhance spring CO2
uptake and annual cycle amplitudes over the Arctic. This effect may be
counteracted by the fact that early growing season onset may lead to growing
season moisture deficits that reduce terrestrial productivity later in the
growing season . Model evidence suggests that climate-driven shifts
in vegetation cover can also enhance GPP. showed
that the interaction of vegetation dynamics and climate change leads to increased GPP
over NH boreal and Arctic regions that in turn drive the observed increases
in NH high-latitude seasonal CO2 amplitudes.
In short, there are several potential drivers of the CO2 annual cycle
amplification that feed back into the climate system. Despite the
representation of these mechanisms by ESMs,
showed that none of the CMIP5 carbon cycle models were able to simulate the
magnitude of the observed increase in atmospheric CO2 seasonality. Since
understanding the drivers of the CO2 seasonality is crucial for model
development, we used the Community Earth System Model (CESM) to study the
contribution of natural drivers of variability in CO2 fluxes to the
increasing amplitude by separating the effects of CO2 radiative forcing
(climate change), CO2 fertilization and N deposition, and land use change
on the atmospheric CO2 annual cycle amplitude over the NH subtropics, NH
midlatitudes, and the NH high latitudes before and after 2100 in an extension of the
high-emission RCP8.5 scenario .
In addition to revealing potential effects of continued increases in CO2
emissions, anthropogenic nitrogen, and land use change up to 2100, extending
the RCP8.5 scenario to 2300 allowed us to assess the behavior of the mean
annual CO2 cycle in a warmer climate following stabilization of
atmospheric CO2 mole fraction and a shift in the terrestrial biosphere
from a CO2 sink to a source as shown by
.
The questions guiding our analysis of CESM extended concentration pathway
simulations are as follows:
Does the relative importance of drivers of the CO2 amplitude trend change after 2100?
For example, do we see evidence of saturation of the CO2 fertilization
effect or evidence of a climatic tipping point after which the CO2
amplitude declines?
Do the regional contributions to CO2 mean annual cycle trends change in response to large changes in climate?
Does the CO2 annual cycle amplitude scale with the hemispheric carbon sink from net ecosystem productivity (NEP) as climate and atmospheric conditions evolve in the future?
The CESM provides a unique platform for exploring these questions in that it
is one of the few prognostic ESMs to include coupled carbon–nitrogen
biogeochemistry and diagnostic atmospheric CO2 variability. This paper is
organized as follows: first, we discuss the ability of the CESM to capture
observed changes in the mean CO2 annual cycle amplitude throughout the NH.
Second, we describe how climate change, CO2 fertilization and
N deposition, and land use change impact the NH CO2 annual cycle amplitude
in the CESM before and after 2100. Third, we examine how forcing from
different regions contributes to the amplitude changes attributed to each
driver. Finally, we discuss our results and provide recommendations for
future analysis.
Methods
Model
We analyzed simulations from the Community Earth System model with coupled
biogeochemistry (CESM1(BGC); ) to explore the
role of environmental change on land–atmosphere carbon exchange. The
Community Atmosphere Model (CAM, version 4; ) and
the Community Land Model (CLM, version 4; ) were
the most important components for our research, but all components of the
model, including physical and biogeochemical ocean processes and sea ice
processes, were interactive in the model configuration. The CAM4 was run on a
0.94∘ × 1.25∘ finite volume grid with 26 vertical
levels. The model simulated climate conditions, including temperature,
precipitation, and humidity, that provide important boundary conditions for
land biogeochemistry. Moreover, the CAM4 directly simulates three-dimensional
transport of atmospheric CO2, as well as separate CO2 tracers derived
from fossil fuel emissions, land exchange, and ocean exchange.
The CLM4 exchanged fluxes of sensible and latent heat, momentum, moisture,
radiation, and terrestrial carbon with the CAM4 and was run at the same
horizontal resolution. Biogeochemistry is represented in CLM4 by a prognostic
carbon–nitrogen model (CLM4CN, ) and fire
model adapted from the model. We note that
important high-latitude processes, such as permafrost carbon dynamics, were
not simulated in the CLM4, meaning that the model may have underestimated
both the seasonal dynamics of soil carbon fluxes and the long-term dynamics
of permafrost melt and the subsequent radiative feedback into the climate
system . In our analysis, we used the CLM4 NEP, defined as the difference between GPP and total
respiration (autotrophic + heterotrophic), to calculate the atmospheric
CO2 annual cycle amplitudes described in Sect. .
Experiments
Three CESM simulations were run from 1850 to 2300 to separate the effects of
climate change, CO2 fertilization and N deposition, and land use change.
The mole fraction of CO2 in the atmosphere is prescribed according to the
RCP8.5 and ECP8.5 scenarios described by , and it
is this value that controls both radiative forcing and CO2
fertilization. However, the CESM retains a separate, spatially varying CO2
tracer that is a diagnostic passive tracer of land, ocean, and fossil fuel
carbon fluxes; the additional carbon exported from the land and ocean to the
atmosphere does not exert any radiative forcing on the climate.
Map of stations where CO2 from the pulse-response code is
computed. Table lists the station identifiers and
locations. The GEOS-Chem regions defined in the pulse-response code (see
Table ) are shaded and grouped according the Arctic
(blue shades), boreal (orange shades), NH temperate (green shades), NH
subtropical (purple shades), tropical (red shades), and SH land (yellow
shades) ecoclimate regions. Greenland and Antarctica (gray shades) were
excluded from the analysis. Gray circles indicate a subset of stations from
which the atmospheric annual CO2 amplitudes were computed from 1985 to 2013
monthly mean observations (bold stations in Table ).
The degree of coupling between CO2 biogeochemistry and radiative forcing
differed across the three runs. In the first simulation, denoted
FullyCoupled, the imposed CO2 was radiatively active, and additional
anthropogenic radiative forcing resulted from prescribed CH4,
chlorofluorocarbons, ozone, and aerosols. In this simulation, the increasing
CO2 was also biogeochemically active, meaning it contributed to CO2
fertilization. Transient land use change (LUC) from agriculture and wood
harvest, and land and ocean N deposition were applied
through 2100, then held at 2100 values through 2300.
and
provide additional descriptions of
the model configuration and analyses of the FullyCoupled simulation during
the 20th century. In the second simulation (NoRad), radiative forcing from
CO2 and other species was fixed at 1850 values, but the changing CO2
mole fraction interacted with biogeochemistry via CO2 fertilization. LUC
and N deposition were likewise prescribed as in FullyCoupled.
also details the design of the
FullyCoupled and NoRad (referred to as “NoCO2Forcing”) simulations
through 2300. We isolated the impact of climate change on the mean annual
CO2 cycle by taking the difference between the FullyCoupled and NoRad
simulations. The third simulation (NoLUC), was configured identically to
FullyCoupled with the exception that LUC was held constant at 1850 values;
therefore, LUC effects on terrestrial carbon exchange were determined from
the difference between FullyCoupled and NoLUC.
Variations in fractional coverage, albedo, nutrient limitations, and surface
energy fluxes among trees, grasses, and crops may enhance or oppose the
effects of climate change and CO2 fertilization on the atmospheric CO2
mean annual cycle amplitude. These changes based on plant functional type (PFT)
were approximated by prescribing transient land cover change through 2100 in
FullyCoupled and NoRad based on annual fractional transition among primary
vegetation, secondary vegetation, pasture (grazing land), and crops described
by , with CESM PFTs detailed in
. The crop model was inactive in the
CESM simulations, and the crop PFT in data
was specified as unmanaged grass .
Therefore, our CESM results do not include anthropogenic influences on CO2
seasonality or agricultural intensification. Moreover, these simulations were
run without dynamic vegetation, which potentially damps feedbacks that could
contribute to changes in the CO2 annual cycle through 2300.
Latitude band, station location, station ID, latitude, and longitude
of CO2 sample locations. Observations from bold stations were analyzed
over the 1985–2013 period.
Latitude band
Station location
Station ID
Latitude
Longitude
NH high latitudes
Alert, Nunavut
ALT
82.45
297.49
Ny-Ålesund, Svalbard
ZEP
78.90
11.90
Barrow, Alaska
BRW
71.30
203.40
NH midlatitudes
Baltic Sea
BAL
55.35
17.22
Shemya Island, Alaska
SHM
52.70
174.10
Hohenpeissenberg, Germany
HPB
47.80
11.02
Hegyhátsál, Hungary
HUN
46.95
16.65
Ulaan-Uul, Mongolia
UUM
44.45
111.10
Trinidad Head, California
THD
41.10
235.80
Shangdianzi, China
SDZ
40.65
117.12
NH subtropics
Tae-ahn Peninsula, South Korea
TAP
36.70
126.10
Mt. Waliguan, China
WLG
36.29
100.90
Lampedusa, Italy
LMP
35.52
12.62
Tudor Hill, Bermuda
BMW
32.30
295.10
Weizmann Institute of Science Station, Negev Desert, Israel
WIS
29.97
35.06
Izana, Tenerife, Canary Islands
IZO
28.31
343.50
Sand Island, Midway, USA
MID
28.21
182.62
Key Biscayne, Florida
KEY
25.67
279.84
Lulin, Taiwan
LLN
23.47
120.87
NH tropics
Mauna Loa, Hawaii
MLO
19.50
204.42
Pacific Ocean (15∘ N)
POCN15
15.00
215.00
Mariana Islands, Guam
GMI
13.39
144.66
Ragged Point, Barbados
RPB
13.17
300.57
Pacific Ocean (10∘ N)
POCN10
10.00
211.00
Christmas Island, Rep. Kiribati
CHR
1.700
202.85
SH tropics
Bukit Kototabang, Indonesia
BKT
-0.20
100.32
Mahé Island, Seychelles
SEY
-4.68
55.53
Maxaranguape, Brazil
NAT
-5.52
324.74
Ascension Island, UK
ASC
-7.97
345.60
Pacific Ocean (10∘ S)
POCS10
-10.00
199.00
Tutuila, American Samoa
SMO
-14.25
189.44
Pacific Ocean (20∘ S)
POCS20
-20.00
186.00
SH
Gobabeb, Namibia
NMB
-23.58
15.03
Easter Island, Chile
EIC
-27.16
250.57
Pacific Ocean (35∘ S)
POCS35
-35.00
180.00
Cape Grim, Tasmania, Australia
CGO
-40.68
144.69
Baring Head Station, NZ
BHD
-41.41
174.87
Palmer Station, Antarctica
PSA
-64.92
296.00
Syowa Station, Antarctica
SYO
-69.01
39.59
Halley Station, Antarctica
HBA
-75.61
333.79
South Pole, Antarctica
SPO
-89.90
335.20
Mapping atmospheric CO2 from surface fluxes
Although the CESM simulated the three-dimensional structure of atmospheric
CO2, we used a pulse-response transport operator to separate imprints of
CO2 fluxes from different regions on the atmospheric CO2 variations.
The transport operator was developed using the GEOS-Chem transport model
(version 9.1.2, ). GEOS-Chem was configured as
in on a 4∘ × 5∘
horizontal grid with 47 vertical layers and was forced with meteorological fields
from the 3–6 hourly Modern Era Retrospective-Analysis for Research and
Applications (MERRA) reanalysis data set . A
tagged 1 Pg C month-1 pulse was released for each of the 20
terrestrial source regions in Fig. for each calendar month
and was allowed to decay for 60 subsequent months. Each 1 Pg C month-1
pulse was distributed spatially according to monthly fluxes from the
Carnegie–Ames–Stanford approach (CASA; see list of acronyms in Table 2) from
.
At a given location, the magnitude and phasing of the atmospheric CO2
response of the pulse depends on the characteristics of atmospheric transport
(Fig. ). For example, at Barrow (BRW) in northern Alaska,
a 1 Pg pulse released in boreal North America (NBNA) in the winter months
(December–January) has a large impact on atmospheric CO2 during the first
1–2 months after a pulse is released (2 ppm, Fig. a),
but more vigorous vertical mixing in the summer months reduces the imprint to
0.5 ppm. In contrast, when the pulse is released from temperate North
America (ETNA, WTNA), there is a phase lag of 2–3 months
(Fig. b, c), and when the pulse is released from the
Amazon (AMZN), there is a delay in the peak response at BRW of at least 4 months (Fig. d). Following the 12-month period in which
pulses were released, the signals were allowed to decay for 60 subsequent
months, at which point CO2 was well-mixed in the atmosphere
(Fig. a–d). We then sampled GEOS-Chem at the locations
of 41 NOAA cooperative CO2 flask sample sites (;
Table , Fig. ) for each of the 72 total
months simulated. This resulted in a CO2 transport operator matrix with
the dimensions Nreg.×Nobs.×Nmon..
The imprints of 1 PgC pulses emitted in individual months (colored
curves) on atmospheric CO2 at Barrow (BRW). Imprints on the atmosphere sampled at BRW are shown from
selected regions: (a) northern boreal North America (NBNA), (b) eastern
temperate North America (ETNA), (c) western temperate North America (WTNA),
and (d) the Amazon (AMZN). For clarity,
we plot only the first calendar year (months 13–24) after the pulses were
released.
Mean annual cycles of atmospheric CO2 derived from (blue curves)
NEE run through the pulse-response function and (black curves) the CESM land
CO2 tracer for (a–d) Barrow (BRW), (e–h) Shemya Island (SHM), (i–l) Key
Biscayne (KEY), and (m–p) Mauna Loa (MLO) in 1990–1999, 2090–2099,
2190–2199, and 2290–2299.
We used monthly mean NEP from the CESM to derive atmospheric CO2 from the
pulse-response function. We aggregated NEP fluxes from CLM4 to the spatial
scale of the 20 source regions (Fig. ) and used matrix
multiplication to propagate these fluxes to atmospheric CO2. We calculated
the monthly mean CO2 mole fraction at the observation sites (e.g., blue
lines in Fig. ) by summing over the instantaneous
contributions from all regions and the background contributions from fluxes
released during the 60 previous months to get a CO2 response matrix with
the dimensions Nobs.×Nmon.. We analyzed both the
CO2 fields from global fluxes and the CO2 patterns influenced only by
larger regions representing Arctic, boreal, temperate, subtropical, tropical,
and Southern Hemisphere (SH) ecosystems. We calculated the CO2 annual
cycle amplitude values as the peak-to-trough differences in CO2 summed
over each component region (e.g., the CO2 annual cycle amplitude at a
given station from pulses emitted from the Arctic was calculated as the
peak-to-trough difference in the sum of CO2 from pulses emitted by the
blue regions in Fig. ). We note that our analysis focuses on
surface observations of atmospheric CO2 and does not include aircraft
measurements.
The advantage of the pulse-response method is that we can efficiently compute
the regional contribution to changes in atmospheric CO2; it would be
prohibitively expensive to run a full atmospheric transport model for each of
the regions separately for 350 years. However, using this simplified
transport operator introduces errors. To evaluate the pulse-response method,
we show a comparison in which we have generated CO2 using net ecosystem
exchange (NEE), which includes fire, harvest, and land use fluxes
(Fig. ), since the land CO2 tracer in the CAM4 is
derived from NEE (despite using NEP for subsequent analyses). The
errors are generally less than 2 ppm between the full transport and pulse-response calculations due to different model boundary layer schemes and
atmospheric transport (Fig. c). We note that the largest
differences were during the last century of the simulation, which was likely
due to shifts in atmospheric transport in response to the dramatic climate
change in the CAM4. The fact that long-term trends in transport are not
simulated by the pulse-response approach is one of the major sources of bias.
In a site-by-site comparison (Fig. ), the increasing bias
through 2300 appears to be due to amplification of existing biases in the
pulse-CO2 compared to the full transport CO2. A second source of
uncertainty is that the spatial distribution of fluxes within each region is
different in CLM compared to CASA. We expect that this has a minimal impact
based on results from , who showed that a
similar pulse-response code using different transport models did a reasonable
job (r2=0.8) of simulating the fossil fuel influence on CO2 despite
that fossil fuel emissions show a vastly different spatial configuration than
do ecosystem fluxes. In our analysis, we aggregate the sites into high-,
mid-, subtropical, and tropical latitude belts to minimize local effects at
individual sites and instead to focus on large-scale trends owing to broad
patterns of changing fluxes.
CO2 annual cycle amplitudes in the FullyCoupled simulation
derived from (a) the CESM land CO2 tracer and (b) running NEE from the
CLM4 through the pulse-response code. (c) The difference between the land
tracer and NEE-derived CO2 annual cycle amplitudes.
We assessed the validity of ignoring ocean contributions to trends in the mean annual cycle of CO2 by calculating the
CO2 amplitudes in the CAM land and ocean tracers. We found that the
contemporary peak-to-trough amplitude in the ocean tracer averaged across our
high-latitude stations was 2 ppm (in contrast to 10 ppm in the land
tracer). Although both the land and ocean amplitudes grow with time, by 2300
the high-latitude ocean tracer had an amplitude of 3 ppm, only 18 % of
the land amplitude for this time period.
Atmospheric CO2 time series analysis
To place the CO2 annual cycle amplitudes simulated by the CESM in the
context of observations, we quantified observed and simulated CO2 annual
cycle amplitude at NOAA observatories before aggregating amplitudes across
four latitude bands spanning 60–90∘ N (NH high latitudes),
40–60∘ N (NH midlatitudes), 20–40∘ N (NH subtropics), and
0–20∘ N (NH tropics). We identified a subset of stations
in the NOAA Global Monitoring Division and
Scripps Institute of Oceanography networks
with better than 95 % temporal coverage of monthly mean values from
1985 to 2013 (gray circles in Fig. ). The trends at these
stations were calculated iteratively as a second-order polynomial, as
described by . After subtracting the
trend from the raw observations, we calculated the peak-to-trough amplitude
(AObs) for each calendar year in which observations existed. We
then aggregated Aobs from all stations within the specified
latitude bands to determine a regionally averaged amplitude.
We calculated the regional CO2 amplitudes for the FullyCoupled
simulation (AFC) using a nearly identical methodology. However,
due to the length of the simulated time series, we detrended the data in
10-year increments. For CESM output, we used only the sampling locations
with greater than 95 % temporal coverage for comparison with the
observations (Fig. , gray circles) but aggregated
amplitudes at a larger set of marine boundary layer observatories when
assessing future trends (Fig. , black circles). Due to the
flexible transport operator, we separately calculated amplitudes from NoRad
(ANoRad) and NoLUC (ANoLUC) simulations and were
further able to simulate only the contribution from specified ecosystem
types. The contribution of climate change to the CO2 mean annual cycle
amplitude (AClim) was calculated from the difference between
AFC and ANoRad. Likewise, the LUC contribution to the
annual cycle amplitude was calculated from the difference between
AFC and ANoLUC.
Discussion and conclusions
CO2 annual cycle amplitude trends and applications for model evaluation
By analyzing CESM simulations run to 2300 with the ECP boundary conditions,
we identified notable carbon cycle interactions that were not apparent before
2100, the nominal end date for CMIP5 runs. We found that the mean NH
atmospheric CO2 annual cycle amplitude increased by 65 % from 1950 to
2100 and by an additional 30 % by 2300 in the CESM1(BGC). Despite
significant changes in climate and atmospheric CO2 mole fractions in the
extended simulation, the sensitivity of the CO2 annual cycle amplitude to
climate and fertilization drivers was remarkably similar both before and
after 2100. We likewise found that regional contributions to the NH CO2
seasonal amplitude trend were generally consistent throughout the simulation,
in response to the second question posed in the introduction. CO2
fertilization was the dominant driver of CO2 annual cycle amplification
for most of the NH, with the notable exception of Arctic ecosystems where
temperature increases drove amplification prior to 2100. GPP leveled off in
boreal high latitudes when annual mean temperatures exceeded 284 K, which
contributed to small amplitude declines in the 23rd century. CO2
fertilization increased the CO2 annual cycle amplitude globally for the
duration of the simulation even after the growth rate of CO2 slowed during
the 23rd century, suggesting that the CO2 fertilization effect had not
saturated in the CESM when CO2 mole fractions were around 2000 ppm.
We saw evidence of hysteresis in the relationship between carbon cycle
diagnostics that was only apparent after 2100, with respect to our third
research question. For example, we found that the relationship between the NH
mean CO2 annual cycle amplitude and the NH net land carbon sink changed
after 2100. A strong positive relationship between the CO2 annual cycle
amplitude and terrestrial carbon sink, which was previously noted by
, became decoupled in the mid-22nd century when sink
strength began to decrease after 2100, while the CO2 annual cycle
amplitude continued to increase. When the NH CO2 annual cycle amplitude
began to decrease in the 23rd century, the correlation between the amplitude
and land carbon sink weakened considerably compared to the relationship
between the two quantities during the 20th and 21st century correlations.
Our analysis of extended concentration pathway simulations suggests several
approaches for how to leverage multi-decadal atmospheric CO2 observations
for model evaluation. For example, consideration of only the trend in the
hemispheric mean annual cycle, rather than trends in the latitudinally
resolved mean annual cycle of CO2, may obscure model biases. The Extended
Concentration Pathway simulations run in the CESM with coupled
biogeochemistry show that the NH mean annual cycle amplitude of atmospheric
CO2 increased by 16 % from 1985 to 2013. The relative increase was in
line with the observed 24 % increase over the NH during the same time
period. However, the spatial pattern of the relative amplitude change was
more uniform throughout the NH than observed. Furthermore, the trend in the
absolute magnitude of the amplitude at high latitudes was about half of the
observed trend (0.05 vs. 0.09 ppm yr-1). This result highlights the
importance of considering meridionally resolved atmospheric CO2 data that
explicitly account for the role of transport since analysis of only
hemispheric spatial patterns obscures incorrect spatial patterns simulated by
the CESM. Analysis of the long-term CESM simulations are suggestive of
patterns we might see in ongoing monitoring of the CO2 annual cycle
amplitude. CESM simulations show that the major drivers of the mean annual
cycle amplification impact differential imprints on atmospheric CO2 in
different latitude bands. For example, CO2 fertilization leaves the
largest imprint in both absolute and relative terms on midlatitude CO2,
whereas climate change may amplify high-latitude CO2 while having a
near-neutral impact on CO2 annual cycle amplitudes south of 60∘ N
(Fig. ). These fingerprints may be useful for developing
hypotheses regarding observed trends and determining future observational
strategies to monitor carbon–climate feedbacks.
Several recent papers have considered how the amplitude of NH net carbon
exchange has changed over the historical period in different categories of
prognostic models. analyzed MsTMIP terrestrial
ecosystem models to determine how atmospheric CO2, climate change, and
land use affect the NH flux amplitude for the historical period, and
analyzed the net terrestrial flux to the atmosphere in
TRENDY models. Both of these studies find that CO2 fertilization is the
strongest driver of increasing ecosystem productivity and thus the amplitude
of the net carbon exchange in the NH, consistent with our results. A
significant difference between the approach used by these papers and our
study is that they consider the net flux amplitude, whereas we propagate
fluxes using an atmospheric transport operator to determine the influence on
latitudinally resolved atmospheric CO2 fields. Given the importance of
atmospheric transport to the mean annual cycle of atmospheric CO2
e.g.,, and the small biases induced by the
simplified pulse-response transport operator, we recommend that future
studies explicitly simulate the full atmospheric CO2 field.
Uncertainties and future model needs
The mean annual cycle of atmospheric CO2 is a first-order diagnostic of
terrestrial carbon exchange and its trend with time integrates a range of
environmental and human factors . An
active area of carbon cycle research is determining the extent to which
coupled ESMs provide predictive skill for future carbon–climate feedbacks. We
note that many of the methods used to evaluate the carbon cycle in ESMs rely
on benchmarking short-term responses to either seasonal or interannual
climate variability , or they rely on
extrapolating future behavior based on some mechanistic link between
short-term and long-term variability . The changing CO2 annual cycle provides a unique
opportunity to gauge a model's sensitivity to slow-varying climate and
environmental changes since we observed large trends in this quantity
over the instrumental record .
However, biases in seasonality in the CESM1(BGC) that lead to smaller
increases in NH atmospheric CO2 seasonal amplitudes in the CESM compared
to observations during 1950–2010 prompt further model development. Moreover,
the relatively smooth response to extreme changes in temperature and CO2
suggests that the CESM may not parameterize processes that could cause
nonlinear carbon cycle feedbacks. CESM1(BGC) did not include
paramaterizations for permafrost carbon dynamics, which have since been
improved in CESM . The lack of permafrost
dynamics likely has a large impact on CO2 annual cycle trends, especially
later in the simulation when global mean temperature has increased by over
10 K in the FullyCoupled simulation. Thus, short soil carbon turnover time
in CLM4 may have contributed to the amplitude underestimation by damping
ecosystem respiration outside of the growing season
and would affect
both baseline values and trends. Ongoing model development in the CESM
includes improved representation of permafrost carbon
, and thus future model configurations will
provide an improved tool for investigating a process that may provide one of
the tipping points we identified in our key science questions.
In addition, found that interaction between
climate change and changes in vegetation cover over northern high latitudes
was the primary driver of the north–south gradient in observed NH atmospheric
CO2 seasonal amplitude trends, indicating that the lack of dynamic
vegetation in the CLM4 likely contributes to underestimation of the seasonal
amplitudes by the CESM. Tree cover is expected to expand further northward
with climate change e.g.,, which may
contribute to the long-term increase in NEP flux amplitude within high-latitude ecosystems. In contrast, drying at lower latitudes may lead to
replacement of trees with grasses and subsequent decreases in NEP amplitude.
An ecosystem demography version (CLM-ED) that will permit successional
patterns in response to environmental change is presently under development.
We consider the documentation of trends in the static-vegetation
configuration presented in this paper to be a crucial first step toward
eventually determining the sensitivity of land–atmosphere biogeochemical
couplings in more sophisticated future configurations of the CESM model.
Development is also underway to represent irrigation and fertilization in
croplands in future versions of the CLM. and
suggest that agricultural amplification,
facilitated by irrigation and fertilization, may be an important driver of
the observed mean annual cycle trend. In the CESM, however, crop cover is
currently treated as unmanaged grass. Thus, these agricultural practices are
not explicitly modeled and do not mitigate the reduction in tree cover in
the FullyCoupled simulation. Our results indicate that explicit consideration
of human modifications may be necessary for prognostic models both to match
observations and to provide realistic predictions of future changes. After
accounting for land management contributions to the amplitude increase, the
sensitivity of the CO2 amplitude to natural factors may be reasonable. Our
results suggest that model development focused on human modification of
carbon fluxes (e.g., by agriculture, , or by
disturbance, ) may facilitate improved comparison
both of mean behavior and trends.