Introduction
Climate models are continuously evolving to include more processes and
interactions at higher resolutions and their number has increased rapidly in
recent years. In addition, a number of institutes worldwide have been
developing Earth system models (ESMs), which are able to simulate both
physical and biogeochemical processes through the inclusion of the land and
ocean carbon cycles.
The evaluation of ESMs in terms of their capability to reproduce climate and
carbon-related variables over the historical period (i.e. 1850–2005) is
crucial prior to using such models for future predictions. Comparisons are
usually performed with observation-based products, if available, but also
with other ESMs to identify common weaknesses.
The performance of 18 ESMs that participated in the Coupled Model
Intercomparison Project phase 5 (CMIP5) has been evaluated
in for the present-day climate. They found that all models
correctly reproduce the main climate variables controlling the spatial and
temporal variability of the carbon cycle. However, large differences exist
when reproducing specific fields. In terms of the land carbon cycle, an
overestimation of photosynthesis and leaf area index (LAI) was found for most
of the models. In contrast, for the ocean an underestimation of the net
primary production (NPP) was noted for a number of models.
also found significant regional variations in model performance.
Eight of these CMIP5 ESMs were also evaluated in ,
highlighting that temporal correlations between annual-mean carbon cycle and
climate variables vary substantially among the eight models. Large inter-model
disagreements were found for NPP and heterotrophic respiration (Rh). In
agreement with , also noted that the CMIP5
historical simulations tend to overestimate photosynthesis and LAI.
compared and evaluated 11 CMIP5 ESMs in terms of their
variations in soil carbon. The correct representation of soil carbon in the
model is important in order to accurately predict future climate–carbon
feedbacks. Soil carbon simulations of the 11 models were compared against
empirical data from the Harmonized World Soil Database (HWSD) and from the
Northern Circumpolar Soil Carbon Database (NCSCD). A large spread across all
models was found (nearly 6 fold) and the spatial distribution of soil carbon,
especially in the northern latitudes was found to be poor in comparison to
HWSD and NCSCD, which means that most ESMs were poorly representing
grid-scale soil carbon.
showed that CMIP5 models appeared to capture the
observed pattern of anthropogenic carbon storage in the ocean, particularly
in the Southern Ocean. However, overall they underestimate the magnitude of
the observed oceanic global anthropogenic carbon storage since the
pre-industrial.
The representation of the global carbon cycle in ESMs continues to be
challenging. For example, large uncertainties exist for the climate–carbon
feedback, which can be mainly attributed to terrestrial carbon cycle
components . Terrestrial ecosystem models
show large variations when driven with future climate scenarios
due to differences in model formulation and
uncertainties in process parameters .
The Australian Community Climate and Earth System Simulator (ACCESS)
participated in CMIP5, but in a climate-model-only version. A selection of
CMIP5 simulations have now been performed with the ESM version of ACCESS,
ACCESS-ESM1 . Here, we present the performance of the land and
ocean carbon cycle components of ACCESS-ESM1 over the historical period
(1850–2005). First, we briefly assess ACCESS-ESM1 simulation of climate
variables that are relevant to the carbon cycle (Sect. ). We
then focus on the response of the carbon cycle to the historical forcing
(Sect. ) and comparison of various present-day simulated
carbon variables with observations (Sect. ).
provides complementary analysis of the ACCESS-ESM1 pre-industrial simulation.
Model configuration, simulations and comparison data
Historical simulations (Sect. ) are performed with two
model configurations (Sect. ) and the results compared with
other CMIP5 ESMs (Sect. ) and a number of observed data
products (Sect. ).
Model configuration
ACCESS-ESM1 is based on the ACCESS climate model , but with the
addition of biogeochemical components for ocean and land as described in
Part 1 of this paper . The climate model version underlying
the ESM version is ACCESS1.4, a minor update of the ACCESS1.3 version
submitted to CMIP5 . The relationship between the
ACCESS1.3, ACCESS1.4 and ACCESS-ESM1 versions is illustrated in Fig.
1. also showed that the climate simulations of the
three model versions are very similar.
For the ACCESS-ESM1 version, ocean carbon fluxes are simulated by the World
Ocean Model of Biogeochemistry And Trophic dynamics (WOMBAT)
and land carbon fluxes are simulated by the Community Atmosphere Biosphere
Land Exchange (CABLE) model , which optionally
includes nutrient limitation (nitrogen and phosphorus) for the terrestrial
biosphere through its biogeochemical module, denoted CASA-CNP
. This capability is important because nitrogen, phosphorus
and carbon biogeochemical cycles are strongly coupled, and it has been
demonstrated that nutrient limitation has a large impact on the productivity
of terrestrial ecosystems . Consequently,
global land carbon uptake can be altered significantly. Here we run CASA-CNP
in “CNP” mode with both nitrogen and phosphorus limitation active. This
differentiates the ACCESS-ESM1 simulations presented here from other ESM
simulations for CMIP5, few of which included nitrogen and none of which
included phosphorus.
The CMIP5 models used to assess the ocean response of ACCESS-ESM1
over the historical period in the study. References for all models are
provided in .
Model name
Institute ID
Modelling group
CanESM2
CCCMA
Canadian Centre for Climate Modelling and Analysis
HadGEM-ES
MOHC
Met Office Hadley Centre (additional HadGEM2-ES
(additional realisations by INPE)
realisations contributed by Instituto Nacional de Pesquisas Espaciais)
GFDL-ESM2M
NOAA GFDL
NOAA Geophysical Fluid Dynamics Laboratory
ISPL-CM5A-LR
IPSL
Institut Pierre-Simon Laplace
IPSL-CM5A-MR
IPSL
Institut Pierre-Simon Laplace
MPI-ESM-LR
MPI-M
Max-Planck-Institut für Meteorologie
(Max Planck Institute for Meteorology)
As in , two model configurations are used, differing in their
treatment of LAI. LAI is an important variable in climate
models for describing the biophysical and biogeochemical properties of the
land cover and in CABLE it can either be prescribed or simulated. When
prescribed, monthly values based on MODIS observations are read in through an
external file Sect. 3.1.1. The dataset used here is
limited by having no inter-annual or longer timescale variability.
Additionally, the same LAI is assigned to all plant functional types (PFTs)
within a grid cell even though CABLE simulates multiple PFTs per grid cell.
With prescribed LAI there is no coupling between the LAI and the leaf carbon
pool, which means that vegetation feedbacks cannot be included. These
limitations are removed by making LAI a prognostic variable with the LAI
dependent on the simulated size of the leaf carbon pool. However, if the leaf
carbon pool is not well simulated then this would lead to a poor LAI
simulation with consequent impacts for the climate simulation.
Simulations
All experiments are set up as concentration-driven
simulations, which means that (historical) atmospheric CO2 concentrations
are prescribed as an input to ACCESS-ESM1 and changes in the land and ocean
carbon pools do not feed back on to atmospheric CO2 concentrations
following CMIP5 protocols .
As noted above we run ACCESS-ESM1 in two configurations, with prescribed LAI
(PresLAI) and prognostic LAI (ProgLAI). For PresLAI, the carbon cycle has no
impact on the simulated climate, whereas for ProgLAI there is a small impact
on the climate through biogeophysical feedbacks related to surface albedo,
evaporation and transpiration Sect. 4.1. The difference in
LAI will also have an impact on the land carbon fluxes, whereas the impact on
the ocean carbon cycle is negligible, and therefore our analysis of the ocean
carbon fluxes focusses only on one scenario (i.e. PresLAI).
Both configurations of ACCESS-ESM1 were run for 1000 years under
pre-industrial climate conditions (year 1850) with the
historical simulations starting from year 800 of these control runs. As noted
in the net carbon fluxes for land and ocean did not
equilibrate to zero. At the end of the control run (i.e. year 800 to 955),
global NEE (net ecosystem exchange) is
0.3 PgCyr-1 for PresLAI and 0.08 PgCyr-1 for
ProgLAI. The net outgassing from the ocean is about 0.6 PgCyr-1
at the end of the control run. We take this drift into account when we
calculate the net uptake of carbon for land and ocean.
The historical simulations use external forcing for 1850–2005, such as
increasing greenhouse gases, aerosols, changes in solar radiation and
volcanic eruptions as used in previous ACCESS versions . For
example, the prescribed atmospheric CO2 increases from 285 ppm
in 1850 to 379 ppm in 2005.
Volcanic eruptions in ACCESS-ESM1 are prescribed based on monthly global-mean
stratospheric volcanic aerosol optical depth , which is then
averaged over four equal-area latitude zones, similar to the way it is done
in the Hadley Centre Global Environmental Model (HadGEM)
. Globally significant volcanoes within the
historical period are Krakatoa (1883), Santa Maria (1903), Agung (1963), El
Chichón (1982) and Pinatubo (1991). Tropospheric aerosols are either
calculated interactively (i.e. sea salt and mineral dust) or are based on
emission datasets (i.e. sulfate and organic carbon) and increase rapidly
from 1950 Fig. 4.
The simulations do not include any land use change; the distribution of PFTs
used in the pre-industrial simulation is used throughout the historical
period.
Comparison with CMIP5 models
ACCESS-ESM1 is compared against other ESMs that participated in CMIP5 and are
available on the Earth System Grid. The models used in this paper are shown
in Table with the references provided in
. As not all years were available for these simulations, we
focussed on the period 1870–2005 and used only the first ensemble member for
each ESM. In assessing the response of the CMIP5 models, we calculated the
median and the 10th and 90th percentiles following . This
allows us to both assess how well ACCESS-ESM1 captures the median and whether
it falls into the range of existing CMIP5 models.
Observations
We use the following observational data products to compare against
ACCESS-ESM1 outputs. Climate variables are assessed, where this is helpful
for interpreting the carbon simulation. For example, the land carbon balance
is mainly controlled by surface temperature and precipitation
, whereas the ocean carbon balance is mainly influenced by
sea surface temperature (SST) and mixed layer depth (MLD)
.
Land surface temperature and precipitation: Climate Research Unit
(CRU) 1901–2013 time series dataset at version 3.22 ,
statistically interpolated to 0.5∘ × 0.5∘ from
monthly observations at meteorological stations across the world's land area
(excluding Antarctica). A low-resolution version at 5∘ for land
surface temperature anomalies (CRUTEM4, ) is used for the
period 1850–1900.
SST: the high-resolution
(1∘ × 1∘) Hadley SST1 in the
period 1870–2006. We also use data from the World Ocean Atlas climatology
WOA2005; in the Taylor diagram.
Climatological MLDs: for the
historical period, based on the density mixed layer criteria of a change
density of 0.03 kgm-3 from the surface.
Ocean NPP: from SeaWIFS calculated with
the VPGM algorithm of .
Global ocean and land carbon flux: Global Carbon Project (GCP)
estimates of annual global carbon budget components and their uncertainties
using a combination of data, algorithms, statistics and model estimates
. The GCP residual land sink is estimated as the
difference of emissions from fossil fuel and cement production, emissions
from land use and land cover change (LULCC), atmospheric CO2 growth
rate and the mean ocean CO2 sink. The 2014 global carbon budget
provides annual values for the period 1959–2013.
Gross primary production (GPP): upscaled data from the flux network
(FLUXNET) using eddy covariance flux data and various diagnostic models
. Gridded data at the global scale is provided by
using a machine learning technique called model tree ensemble
to scale up
FLUXNET observations. Global flux fields are available at a
0.5∘ × 0.5∘ spatial resolution and a monthly
temporal resolution from 1982 to 2008.
LAI: global LAI derived from the third generation (3g) Global
Inventory Modelling and Mapping Studies (GIMMS) normalised difference
vegetation index (NDVI)3g dataset. Neural networks were trained first with
best-quality and significantly post-processed Moderate Resolution Imaging
Spectroradiometer (MODIS) LAI and Very High-Resolution Radiometer (AVHRR)
GIMMS NDVI3g data for the overlapping period (2000–2009) to derive the final
dataset at 1/12∘ resolution and a temporal resolution of 15 days
for the period 1981–2011 .
Soil organic carbon (SOC): the HWSD
represents the most comprehensive and detailed globally
consistent database of soil characteristics that is currently available for
global analysis. We use an upscaled and regridded version of the HWSD with
the area-weighted SOC calculated from the soil organic carbon (%), bulk
density and soil depth .
Salinity, DIC (dissolved inorganic carbon) and alkalinity:
observations for salinity come from the World Ocean Atlas climatology
WOA2005;, while DIC and alkalinity are from GLobal
Ocean Data Analysis Project (GLODAP) .
Sea–air CO2 fluxes: seasonal climatology of
based on the 1∘ × 1∘ global
measurements of oceanic pCO2 of .
Anthropogenic carbon uptake: column inventory estimated from
from GLODAP
.
Atmospheric CO2 concentrations: mean atmospheric
CO2 seasonal cycles derived from NOAA/ESRL flask samples as processed
in the GLOBALVIEW data product. These seasonal cycles
are designed to be representative of background, clean air at any given
location. Here, we assess the seasonal cycle for four locations with an
averaging period of about 20 years for Mace Head (53.33∘ N,
9.90∘ W), about 25 years for Alert (82.45∘ N, 62.52∘ W),
about 35 years for the South Pole (89.98∘ S, 24.80∘ W) and about
40 years for Mauna Loa (19.53∘ N, 155.58∘ W).
Performance evaluation
For climate variables such as land surface temperature and precipitation, we
calculate the model variability index (MVI)
. The model (mod) variability at every grid
point i is compared against the observed (obs) variability and then
averaged over the globe in the following way:
MVI=1n∑i=1nsimodsiobs-siobssimod2,
where s is the standard deviation and n the number of grid cells. Perfect
model – observations agreement would result in an MVI of zero. The
definition of a limit to decide if a model performs well or poor is rather
arbitrary. However, and have used a
threshold of MVI <0.5.
For a number of carbon-related variables, we calculate the inter-annual
variability (IAV), defined as the standard deviation of detrended annual-mean
values.
To assess the performance of the ocean carbon cycle against observations we
use a Taylor diagram . We also apply the same analysis to
archived CMIP5 simulations to benchmark the performance of
ACCESS-ESM1 relative to other CMIP5 models. A Taylor diagram allows us to
summarise the bias, relative variability and correlations of the simulations
with the observations. In the plot, the radial distance of a given simulation
from the origin gives the standard deviation of the simulation normalised by
the standard deviation of the observations. The angle from the x axis
provides the spatial correlation coefficient between the simulations and the
observations. The radial distance from the point marked observations gives a
measure of the root mean squared difference between the simulation and observations
normalised by the standard deviation of the observations. The point's colour
represent the bias in the simulation given as the relative difference in the
globally averaged values between simulation and observations calculated as
(mean_model – mean_observations)/mean_observations; positive values show
the model is overestimating the observed value.
ACCESS-ESM1 climatology
Land temperature and precipitation
Carbon fluxes across the historical period will be directly influenced by
increasing atmospheric CO2 and indirectly influenced by changes in
the climate, driven by the increasing atmospheric CO2 and modulated
by other external forcings, such as anthropogenic and volcanic aerosols. In
addition, each climate simulation generates its own internal variability,
with major modes of climate variability such as the El Niño–Southern
Oscillation (ENSO) known to generate large variability in carbon exchange
between the atmosphere and both the ocean and land .
The evolution of temperature and precipitation in ACCESS-ESM1
(Fig. ) over land shows similar characteristics to
ACCESS1.3 historical simulations as well as those
of ACCESS1.4 (P. Vohlarik, personal communication, 2015). Global land surface
air temperature anomalies (relative to 1901–1930) are shown in
Fig. . Both ACCESS-ESM1 simulation scenarios (PresLAI
and ProgLAI) show similar temperature anomalies over most of the historical
period, being close to the observed anomalies through most of the period
(decadal-mean difference smaller than 0.2 K), apart from the 1940s
where the PresLAI scenario shows a larger negative anomaly (decadal-mean
difference of about 0.37 K), which will be discussed later. From
about 1965 to 2005, anomalies are by up to 0.4 K (decadal-mean
difference) lower than observations for both scenarios. This is attributed by
to a likely overly strong cooling response in ACCESS1.3 to
anthropogenic aerosols, offsetting the warming due to greenhouse gas
increases for which ACCESS1.3 responds similarly to a CMIP5 mean
Figs. 2a, 3a. Strong aerosol cooling is supported by
, who found that ACCESS1.3 showed a large global-mean
aerosol effective radiative forcing over the historical period of -1.56 Wm-2,
which is much larger than the IPCC best estimate (-0.9 Wm-2)
but still within the uncertainty range.
Anomalies (reference period: 1901–1930) for (a) globally
averaged surface air temperature and (b) globally averaged
precipitation for land points only for ACCESS-ESM1 (PresLAI, blue; ProgLAI,
red) and observed CRU (black-dashed; before 1901). Major volcanic eruptions
are marked with dashed lines: Krakatoa (1983), Santa Maria (1903), Mt.
Agung (1963), El Chichón (1982) and Mt. Pinatubo (1991).
The inter-annual variability in temperature is well reproduced by both
ACCESS-ESM1 scenarios, showing an MVI of 0.3 (PresLAI) and 0.4 (ProgLAI) for
the period 1901–2005. According to only a few CMIP5 models
show an MVI of lower than 0.5 (although their calculation is based on present
day, i.e. 1986–2005).
Both ACCESS-ESM1 simulations exhibit cooling following major volcanic
eruptions (marked in Fig. ). At first sight, the
ProgLAI run seems to be more sensitive to volcanic eruptions, showing a
stronger cooling particularly for the two most recent major eruptions, El
Chichón in 1982 and Mt. Pinatubo in 1991. However, this difference might be
due to a different ENSO phase for the two runs at the time of the eruptions.
assessed the temperature impact of Agung, El Chichón and
Pinatubo in three ACCESS1.3 simulations (e.g. their Fig. 7), and mean
temperature anomalies from the two ACCESS-ESM1 simulations lie within or only
slightly outside the ACCESS1.3 ensemble range. It is worth noting that
found that the simulated temperature anomalies from
volcanoes tended to be larger in ACCESS than observed, and this was common
across CMIP5 models.
Differences in the year to year temperature anomalies between the two
ACCESS-ESM1 scenarios are likely due to internal climate variability. For
example, between the years 1940 and 1950, the PresLAI run shows a large
negative temperature anomaly and the ProgLAI run shows a positive anomaly.
The negative anomaly for the PresLAI is probably related to a strong La
Niña event (Nino3 index of -1.2) around the year 1945
(Fig. c), whereas in the ProgLAI case we see a small El
Niño event (Nino3 index of 0.6) around the same time.
The temperature anomalies hide an absolute temperature difference between the
two ACCESS-ESM1 simulations; the ProgLAI scenario produces a slightly warmer
climate (0.56 K difference in mean land surface air temperature
averaged over 1850–2005) than the PresLAI run. This is consistent with the
difference in surface air temperature found for the pre-industrial
simulations Sect. 4.1. As noted in , the
warmer climate can be explained by the difference in LAI, which is generally
higher in the prognostic case. This leads to a lower albedo, especially for
evergreen needleleaf forests during the winter months in the Northern
Hemisphere, and consequently to an increase in absorbed radiation. The
difference in LAI for both scenarios is explored in more detail in
Sect. . Compared to the observations the ACCESS-ESM1 runs show a
cooler land surface air temperature by about 0.5 K for the ProgLAI
scenario and 1.1 K for the PresLAI scenario averaged over 1901–2005.
Precipitation anomalies over the land are presented in
Fig. b. Larger differences in the anomalies for the two
ACCESS-ESM1 simulations can be observed around the years 1870–1880, where
the PresLAI scenario shows a positive anomaly and the ProgLAI scenario shows
a mainly negative anomaly. The difference over the remaining time period for
the two runs is generally small. ACCESS-ESM1 simulations compare well with
observed rainfall anomalies until about 1960 (decadal-mean difference smaller
than 8 mmyr-1), with the exception of the period 1911–1920 for
PresLAI (decadal-mean difference of about 12 mmyr-1) and the
period 1951–1960 for ProgLAI (decadal-mean difference of about
17 mmyr-1). After that, observed anomalies are mostly higher
than the simulation results (decadal-mean difference of up to
41 mmyr-1), a feature also seen in the ACCESS1.3 historical
ensemble Fig. 6a. The comparison of absolute rainfall
for the two ACCESS-ESM1 scenarios suggests a dryer climate (approx. 20
mmyr-1) for the ProgLAI run.
Globally averaged sea surface temperature (K) between 1850 and 2005:
red is ACCESS-ESM1 and black is HadiSST . Major volcanic
eruptions are marked with dashed lines: Krakatoa (1983), Santa Maria (1903),
Mt. Agung (1963), El Chichón (1982) and Mt. Pinatubo (1991).
For precipitation we calculate an MVI of 1.7 (PresLAI) and 1.8 (ProgLAI) for
the period 1901–2005, which suggests that the IAV is not well represented in
ACCESS-ESM1. However, according to none of the CMIP5 models
had an MVI close to the threshold of 0.5. Also note that for the calculation
of the MVI for precipitation we had to exclude 60 land points (mainly coastal
points) due to inconsistencies in the regridding.
A reduction in precipitation can be observed following the eruption of major
volcanoes for both ACCESS-ESM1 scenarios, apart from the 1903 Santa Maria
eruption and the 1982 El Chichón eruption, where the PresLAI scenario does
not show a strong anomaly and the ProgLAI anomaly is likely too late to be
due to the volcano. As for temperature, the precipitation anomalies lie
within or close to the ACCESS1.3 ensemble of anomalies presented by
Fig. 9.
Sea surface temperature and mixed layer depth
To assist in the assessment of responses of the ocean NPP and sea–air
CO2 fluxes, the responses of SST and MLD are first
assessed.
The ocean response from ACCESS-ESM1 is compared with the time series of
HadiSST v1 in Fig. . Here we see, that there
is a warm bias in the early part of the historical period. This warm bias in
ACCESS-ESM1 is the same as reported by over the period
1870–1899 in ACCESS 1.3 (0.26 K). In the period 1870–1970, we see that the
warming of the oceans appears to be less climate sensitive than the
observations. However, by the end of the historical simulation (1970–2005)
we notice that ACCESS-ESM1 captures well the observed response of HadiSST in
the later period.
Differences in sea surface temperature (K) between ACCESS-ESM1 and
HadiSST for (a) February and (b) August.
However, despite little global bias in the latter period we see that the
ACCESS-ESM1 SST response, consistent with ACCESS 1.3 , produces
strong spatial differences from observations. Figure shows clear
spatially coherent differences between ACCESS-ESM1 and observations
(1986–2005). Some of these regions show a strong summer warming bias
(>3 K) in areas such as the high-latitude Southern and Pacific oceans,
while in other regions such as the subtropical Atlantic, a strong cooling
bias is present during the same season. This is in contrast to other regions,
such as the high-latitude North Atlantic, that has a strong year round
warming bias. These biases are broadly consistent with known errors
associated with the UK Met Office Unified Model , which
is employed as the atmospheric model in ACCESS-ESM1. Our SST response is also
broadly consistent with other ESMs, such as HadGEM2 , which also
use the UK Met Office Unified Model.
The magnitude of the inter-annual variability of simulated SST is of similar
magnitude as the observations. In response to large aerosol injections
associated with volcanic eruptions, overlain on Fig. , we see
that the ocean does capture a net cooling, as expected
e.g. and consistent with observations.
Interestingly, the magnitude of the cooling is sometimes less than observed
in HadiSST v1 despite the stronger than observed aerosol response in
ACCESS-ESM1.
Differences in mixed layer depth between ACCESS-ESM1 and
observations for (a, c) February and for
(b, d) August. Panels (e) and (f) show
the percentage difference between and ACCESS-ESM1
calculated as ((OBS – ACCESS-ESM1)/OBS) × 100. The mixed layer is
calculated based on a 0.03 kgm-3 density change from the surface
ocean.
Ocean MLDs are compared with the observations following
, based on more than 880 000 depth profiles from research
ships and ARGO profiles, and based on a 0.03 kgm-3 density
change from the surface. Significant advances in autonomous measurement
platforms have allowed the mixed layer to be increasingly well constrained in
all seasons across the global ocean.
Overall we see in the mid- and lower latitudes that the MLD is deeper than
observed in all seasons (Fig. ). However, the very large values
likely represent the differences in the positions of fronts between the
relatively coarse-resolution model relative to the observations rather than
very large differences . In the higher latitudes winter
mixed layers are well captured by ACCESS-ESM1 (Fig. ). This is
encouraging given that many ocean models tend to underestimate winter MLDs
. Simulating winter mixed layers correctly is
critical for setting interior ocean properties supplying nutrients to the
upper ocean to fuel the biologically active growing season
. However, in contrast to the winter, ACCESS-ESM1 appears
to systematically underestimate MLDs in the high-latitude ocean in summer,
60% (or 30–40 m) in the Southern Ocean, Pacific and Atlantic oceans.
In the Southern Ocean, in particular, the underestimation of summer MLDs is
consistent with and , who showed that most
CMIP5 models underestimate summer MLDs. attributed this to a
lack of vertical mixing in CMIP5 rather than sea surface forcing related to
individual models, this is consistent with , who showed that
these biases are also present in the ocean-only simulations of ACCESS-ESM1.
ACCESS-ESM1 carbon cycle response to historical forcing
The increase in atmospheric CO2 over the historical period is
expected to have a direct impact on both land and ocean carbon fluxes.
Additionally, there may be indirect impacts from the change in climate caused
by the increasing atmospheric CO2. These impacts are explored firstly
for land carbon and then for ocean carbon.
Land carbon response
The direct impact of increasing atmospheric CO2 is seen clearly in
the simulated global land GPP
(Fig. a), with increasing GPP for both simulations. The
ProgLAI case gives the larger increase, with fluxes for the final 10 years of
the simulation being 19 % larger than for the first 10 years, compared to
an increase of 11 % in the PresLAI case. This is due to increasing LAI in
the ProgLAI simulation (Fig. b) compared to the
prescribed LAI, which is annually repeating with no increase. Thus, the PresLAI
case captures only the direct CO2 fertilisation effect of more
efficient photosynthesis per leaf area while the ProgLAI case also allows the
growing leaf biomass to increase the global total assimilation. The
IAV in GPP over the whole historical period for
the ProgLAI run is 2.6 PgCyr-1, considerably larger than in the
PresLAI case (1.7 PgCyr-1), but within the range of other CMIP5
models. We also notice a large decadal variability of global GPP for the
ProgLAI case, which is much weaker in the PresLAI case (1.9 vs.
1.3 PgCyr-1 ). Natural variability of the climate is the main
driver for the IAV in GPP for the PresLAI case. The larger variability in the
ProgLAI case is due to the stronger response to volcanic cooling and climate,
causing an increase in LAI and a positive feedback through increased GPP. In
the PresLAI case, without the LAI feedback, the impact of volcanic cooling is
sometimes largely offset by natural climate variability, for example in the
Pinatubo (1991) case.
Mean carbon (C), nitrogen (N) and phosphorus (P) pools sizes in
petagrams (Pg) for pre-industrial (780–799) and present day (1986–2005).
Historical changes (1850–2005) for C are also shown. Biomass comprises leaf,
wood and root pool.
Pre-industrial
Present day
Historical change C
PresLAI
ProgLAI
PresLAI
ProgLAI
PresLAI
ProgLAI
Pool
C
N
P
C
N
P
C
N
P
C
N
P
ΔC
ΔC
Biomass
611
5.7
0.31
731
6.15
0.33
670
6.2
0.34
807
6.8
0.37
69.5
87.2
Litter
117
0.85
0.04
149
1.02
0.05
126
0.9
0.05
163
1.1
0.06
7.6
12.3
SOC
1034
82
9.6
1187
86.1
11.9
1050
83.4
10.1
1217
88.5
12.6
20.5
37
∑
1762
88.6
10.0
2067
93.3
12.3
1846
90.5
10.5
2187
96.4
13.0
97.6
136.5
The difference between the two simulations is less obvious for the net
ecosystem exchange (Fig. c). NEE is a relatively small
flux that represents the difference between respiration (heterotrophic and
autotrophic) and GPP. In the current set-up of ACCESS-ESM1, we do not include
disturbances such as fire and LULCC, which means that in this case NEE also
represents the net flux of carbon from the land to the atmosphere. Both
simulations generally produce small land sinks over most of the historical
period, with some tendency to an increasing sink from the 1920s, followed by
a possible reduction in the sink from the mid-1990s to 2005. The IAV is
relatively large and similar for both scenarios (1.4 vs.
1.3 PgCyr-1) and likely caused by variations in GPP
that are moderated by respiration, especially in
the ProgLAI case. Table 2 found similar IAV in the
pre-industrial simulation with larger GPP IAV in the ProgLAI case offset by
positively correlated leaf respiration IAV. Decadal variability for the
ProgLAI run is larger than for the PresLAI run (0.7 vs.
0.3 PgCyr-1).
Larger decadal variability in the ProgLAI run can be explained by the
stronger response to volcanic eruptions. In principle, aerosols scatter
incoming solar radiation and therefore have a mainly cooling effect. Hence,
an increase in aerosol emissions leads to a decrease in global temperature,
which in turn increases GPP in the tropics and reduces plant respiration
globally in both cases (PresLAI and ProgLAI) and therefore increases NEE.
However, whereas in the PresLAI case the LAI is kept at a constant level, in
the ProgLAI case the LAI is allowed to increase with the leaf carbon pools
(Fig. b). This leads to a further increase in GPP at
the same time (Fig. a), which further increases NEE in
the ProgLAI case.
Temporal evolution of (a) GPP (PgCyr-1),
(b) LAI and (c) NEE (PgCyr-1). GCP estimates
for NEE are shown for comparison in black for the years 1959–2005.
ACCESS-ESM1 results are shown for PresLAI (blue line) and ProgLAI (red line)
with annual values marked in thin dashed lines and a 5-year-running mean in
heavy solid lines. Major volcanic eruptions are marked with dashed lines:
Krakatoa (1983), Santa Maria (1903), Mt. Agung (1963), El Chichón (1982)
and Mt. Pinatubo (1991).
Due to the fact that during the control run our net carbon flux did not
equilibrate to zero Sect. 4.2.2, we calculate the carbon
uptake for both scenarios by subtracting the mean net flux over the
corresponding part of the control run. We estimate a total uptake of carbon
to the land (using the net ecosystem production (NEP), with NEP=-1×NEE) over the historical period of 98 PgC for the
PresLAI scenario and 137 PgC for the ProgLAI scenario. The increase
in biomass over the historical period is 70 PgC for PresLAI and
87 PgC for ProgLAI, (see also Table ). This is
similar to results from CMIP5 models that also do not consider LULCC. For,
example the Beijing Climate Center Climate System Model (BCC-CSM1.1)
estimates an increase in biomass of about 83 PgC over the historical
period and the Institute of Numerical Mathematics Coupled Model (INM-CM4.0)
reports an increase of about 70 PgC . The increase
in combined soil and litter carbon over the historical period is smaller in
ACCESS-ESM1 (28 PgC for PresLAI and 49 PgC for ProgLAI) than
in the two CMIP5 models without LULCC (64 PgC for both, BCC-CSM1.1
and INM-CM4.0).
We can compare the total carbon uptake (here cumulative NEP) from ACCESS-ESM1
with other models and estimates in two ways.
Comparison against land use emission estimates:
the observation-based cumulative historical land carbon uptake is estimated
to be -11±47 PgC , which suggests an almost neutral
behaviour of the land over that period. Since we do not include disturbances
in our model, we do not expect our simulations to match those results.
However, we can compare our calculated cumulative uptake against estimates of
land use emissions to see if they are in a similar range. For example,
reported land use emissions of 108–188 PgC for
1850–2000, comparable to the ACCESS-ESM1 cumulative uptakes.
Comparison against CMIP5 estimates of cumulative NEP:
simulation results from CMIP5 ESMs that include LULCC provide a large range
for the total carbon uptake. Table 4, for example,
reported the separate contributions of NEP and disturbance to cumulative land
carbon uptake for eight CMIP5 models. While NEP ranges from 24 to 1730
(median 387) PgC and disturbance ranges from 3 to 1729 PgC,
the range for land uptake is smaller with two outlying models (-120 and
211 PgC) and the remainder ranging from -59 to 18 PgC. The
estimates of cumulative NEP from ACCESS-ESM1 are at the low end of the CMIP5
range reported in , possibly due to the inclusion of nitrogen
(N) and phosphorus (P) limitation; found a reduction of
1850–2005 NEP from 210 PgC for a carbon-only simulation to
85 PgC with N and P limitation when using CABLE in a low-resolution
Earth system model.
Comparison of Integrated net primary production
(PgCyr-1) in the period 1850–2005 between CMIP5 and ACCESS-ESM1.
The solid red line represents the integrated carbon uptake in
PgCyr-1 from ACCESS-ESM1, while the green line represents the
median of the CMIP5 model with the range overlain (as shaded area) as the
10th and 90th percentiles. Overlain on this plot are the observed values from
SeaWIFS over the period 1998–2005 in black.
Comparison of sea–air CO2 fluxes (PgCyr-1) in
the period 1850–2005 carbon uptake from ACCESS-ESM1. The solid green line
represents the median of the CMIP5, while the shaded area represents the 10th
and 90th percentiles of the CMIP5 model. Overlain on this is the estimated
sea–air fluxes from the Global Carbon Project in black;
and the timing of major volcano eruptions over the historical period.
Ocean carbon response
Figure shows that, consistent with other CMIP5 models, there is
no statistically significant trend of ocean NPP globally over the historical
period. The global-mean NPP from ACCESS-ESM1 of 51 PgCyr-1 is
close to that calculated from the SeaWIFS data of 50 PgCyr-1 for
1998–2005. Furthermore, it is also in agreement with estimates, based on
observations, of global NPP of between 45 and 50 PgCyr-1
. The ACCESS-ESM1 NPP is larger than the median CMIP5
model value of 37 PgC; however, NPP in CMIP5 models is associated with a very
large range .
The evolution of sea–air CO2 fluxes in the period 1850–2005 is shown
in Fig. . Overlain on this plot is the timing of the major
volcanic eruptions, the estimated sea–air CO2 flux from the GCP and results from the CMIP5 model
archive. We also take into account the drift over the corresponding part of
the control run. Here we see very good agreement with the CMIP5 models in the
period 1870–1960, with the ACCESS-ESM1 sitting close to the median of the
CMIP5 models, and well within the range of the CMIP5 models. After 1960,
ACCESS-ESM1 shows greater uptake than the median of CMIP5 models, and appears
to more closely follow the observed value from the GCP, lying at the 10th
percentile of the CMIP5 range. For 1960–2005, ACCESS-ESM1 gives a mean
sea–air CO2 flux of 1.8±0.1 PgCyr-1 in good
agreement with the estimated GCP value of 1.9±0.3 PgCyr-1,
and larger than the estimate from CMIP5 models of 1.56±0.1 PgCyr-1. For 1986–2005, the sea–air CO2 is
2.2±0.1 PgCyr-1 from ACCESS-ESM1, the same as from the GCP
(2.2±0.2 PgCyr-1), and larger than the median CMIP5 model
value of 1.8±0.1 PgCyr-1. The cumulative uptake of carbon by
air–sea CO2 fluxes in the period 1959–2005 from ACCESS-ESM1 is
83 PgC, which is in good agreement with the GCP value of 82 PgC
over the same period. These results highlight that
ACCESS-ESM1 show good skill at capturing the globally integrated ocean carbon
uptake at the global scale.
Mean annual cycle of GPP (PgCmonth-1) for the period
1986–2005. ACCESS-ESM1 results are shown in blue (PresLAI) and red
(ProgLAI). Observation-based estimates are shown in black.
Evaluation of the present-day carbon cycle
The last 20 years of the historical simulation (1986–2005) is used to
evaluate the simulated carbon cycle against observation-based products.
Analysis considers the land, ocean and atmosphere in turn.
Land carbon
GPP
Both ACCESS-ESM1 runs (PresLAI and ProgLAI) provide a mean GPP of about
130 PgCyr-1 for 1986–2005. The observation-based estimate of
suggests a GPP of about 119 PgCyr-1 for the
same period. Other studies also suggest a global GPP within the same range:
reported an estimate also based on FLUXNET data of 123±8 PgCyr-1 for the period 1998–2005; used
plant traits to constrain parameters of the Farquhar photosynthesis model and
estimated the global GPP for the same period to be 121 PgCyr-1
(95 % confidence interval from 110 to 130 PgCyr-1) and the
IPCC in its AR4 report states a global value of 120 PgC for 1995
. If compared with other CMIP5 Earth system models, which
were divided into two groups by , ACCESS-ESM1 lies in the
middle of the lower group with the range 106 to 140 PgCyr-1. It
was also noted by , that the group of CMIP5 models with a GPP
above 150 PgC did not include nitrogen limitation and might therefore
overestimate GPP. ACCESS-ESM1 contains both nitrogen and phosphorus
limitation, which may provide a more realistic simulation of carbon uptake by
the terrestrial biosphere.
Spatial distribution of (a, c, e) GPP and (b, d, f) GPP IAV (kgCm-2yr-1) for (a, b) PresLAI,
(c, d) ProgLAI and (e, f) observation-based estimates.
A number of studies that base their estimates on observations suggest that a
global GPP of about 120 PgCyr-1 may be somewhat too low. For
example, provided a best guess of
150–175 PgCyr-1 and an estimate of
146±19 PgCyr-1. However, the estimate by is
based on the largest set of observations and also provides a spatial
distribution of GPP. In the following, we therefore use this product for the
validation of the ACCESS-ESM1 land carbon component.
The mean annual cycle of GPP as simulated by the ACCESS-ESM1 is shown in
Fig. for both scenarios as Fig. 8.
Observation-based estimates by are also shown for comparison.
At the global scale both ACCESS-ESM1 runs show a similar behaviour and they
both overestimate GPP by about 2 PgCmonth-1 (peak amplitude) if
compared with the observations as discussed earlier. However, when we split
GPP into its contributions from three latitudinal regions we notice larger
differences between the two ACCESS-ESM1 simulations. The ProgLAI simulation
shows a much more productive northern region (by about
2 PgCmonth-1) and a lower GPP in the tropics (by about
0.2 PgCmonth-1), which compensated for at the global scale.
Overall, both ACCESS-ESM1 simulations show good agreement with the
observations in terms of the amplitude, with only a small bias of up to
2.2 PgCmonth-1 for the globe and the Northern Hemisphere. In
contrast, a large number of CMIP5 models produce a strong positive bias
during June–August on a global scale and for the Northern Hemisphere
. Agreement with observations in terms of the phase is
generally good, accept for the tropics, where ACCESS-ESM1 fails to accurately
reproduce the phase. However, as noted by this is common
amongst CMIP5 models.
Mean annual cycle of LAI for the period 1986–2005. ACCESS-ESM1
results are shown in blue (scenario with prescribed LAI) and red (scenario
with prognostic LAI). Observation-based estimates are shown in black.
The spatial distribution of GPP is presented in Fig.
along with its IAV for the last 20 years of the historical period. Generally
there is good agreement in the spatial pattern of GPP between ACCESS-ESM1
with prescribed LAI and the observation-based estimate (95% of all land
points have errors smaller than 0.5 kgCm-2yr-1). However,
there are some small differences mainly in tropical regions (i.e. central
Africa). The ACCESS-ESM1 ProgLAI run shows a larger GPP in the Northern Hemisphere, mostly in
the boreal regions, and a lower GPP for large parts of South America
(86% of all land points have errors smaller than
0.5 kgCm-2yr-1). Comparing the IAV of GPP for the two
ACCESS-ESM1 runs reveals large differences. Whereas the PresLAI run shows
little variability for most areas, the ProgLAI run shows large hotspots in
South America and south-eastern Australia of up to
0.5 kgCm-2yr-1, which are caused by the LAI feedback as
discussed previously. The observation-based estimate of GPP shows large areas
of variability over the continents, but the distribution and magnitude are
quite different to the ACCESS-ESM1 runs. However, as pointed out in
one of the limitations of the GPP observational product is
the magnitude of the IAV.
LAI
Global LAI estimates are mainly derived from satellite observations and
various products are available. The prescribed LAI in ACCESS-ESM1 is based on
MODIS observations with no IAV. If compared with the
observation-based estimates of , which uses a combination of
MODIS and AVHRR data, over the last 20 years of the historical period (mean
of 1.4), we notice that our current prescribed LAI is somewhat smaller (mean
of 1.3), but agrees well in terms of its seasonal cycle
(Fig. ). There is a number of reasons why remote sensing
LAI products differ from each other, i.e. because different sensors and
algorithms are used .
The prognostic LAI, which is calculated by CASA-CNP, is significantly higher at
the global scale (mean: 1.7) and also shows a different seasonality with its
peak in August, whereas the observations suggest the peak is in July
(Fig. ). In CABLE the phenology phase is currently
prescribed and the leaf onset might be defined as too late for deciduous
vegetation, which leads to a shift in the LAI peak by about 1 month.
The global seasonal cycle of LAI is mainly influenced by the northern
extra-tropics and we notice that leaf coverage throughout the year and
especially in autumn and winter is too high in the ProgLAI case. We clearly
overestimate the mean LAI (observations suggest a mean of 1.3) and
underestimate the seasonal variability. On a PFT level the main contributor
to this is evergreen needle leaf forest, which produces a large value (mean
3.8) over the whole year with only a very small seasonal cycle. In the
tropics we underestimate LAI by a significant amount (mean of 1.5 in
comparison to 2.3 as suggested by observations). This is mainly due to C4
grass showing an LAI, which is about a factor of 5 smaller than the
observations. attributed the low simulated LAI of C4 grass to
a large sensitivity to rainfall and the inability of CABLE to grow back C4
grass after a die back.
The overestimation of the LAI for evergreen needle leaf forest and the
underestimation for C4 grass have a direct impact on GPP, which is also too
large for evergreen needle leaf and too low for C4 grass. In CABLE, the
calculation of GPP is related to APAR (absorbed photosynthetic active
radiation), which is the product of FPAR (fraction of photosynthetically
active radiation) and PAR (photosynthetically active radiation) with FPAR
calculated from the LAI.
At the global scale, most CMIP5 Earth system models also tend to overestimate
LAI Fig. 11, ranging from 1.5 in December–January to
almost 3.5 in June–August. reported that only two models
captured the main feature of the global LAI pattern, whereas the remaining 16
models overestimate the global LAI with some even exceeding a mean of 2.4. At
the regional scale the ACCESS-ESM1 prognostic LAI is within the CMIP5 range
for both hemispheres, but below the CMIP5 range for the tropics.
Spatial distribution of organic soil carbon (kgCm-2)
(a) using prescribed LAI, (b) prognostic LAI and
(c) observation-based estimated from HWSD.
NEE
We compare our NEE results against estimates of the residual land sink from
the GCP for 1959–2005 (Fig. c).
The mean residual land sink and inter-annual variability for this period is
estimated to be about 1.9 ± 1.0 PgCy-1 compared to
1.4 ± 1.3 PgCy-1 for PresLAI and
1.8 ± 1.6 PgCy-1 for ProgLAI. In all cases the IAV is
large relative to the mean uptake, but more so in the ACCESS-ESM1
simulations. The large IAV makes it difficult to be definitive about land
uptake trends over this period, though there is some suggestion of slightly
increasing uptake in the GCP budget estimates but slightly decreasing uptake
in the ACCESS-ESM1 simulations. This might be better assessed using an
ensemble of simulations and extending the analysis closer to 2015 through use
of the RCP (Representative Concentration Pathway) scenario
simulations. Simulations without anthropogenic aerosols would also be useful
to determine whether the relatively strong cooling due to tropospheric
aerosols in ACCESS-ESM1 is impacting the decadal evolution of land carbon
uptake.
CNP pool sizes
The amount of carbon, nitrogen and phosphorus stored in the biomass and soil
of terrestrial ecosystems as simulated by ACCESS-ESM1 is compared against
other estimates from the literature. Here, we refer to the terrestrial
biomass as the sum of living above-ground (leaf and wood) and below-ground
(roots) material. All mean pool sizes and spatial distributions derived from
ACCESS-ESM1 are calculated over the last 20 years of the historical period
(1986–2005).
Carbon pool sizes simulated with ACCESS-ESM1 are in general smaller for the
PresLAI scenario as shown in Table . The total carbon in
the terrestrial biomass amounts to 670 (PresLAI) and 807 PgC (ProgLAI). The
IPCC reports two different estimates of 466 and 654 PgC
for the global plant carbon stock, depending on the data being used. This
would imply that our plant carbon pools are somewhat to large, especially for
the ProgLAI scenario. However, we have to take into account that we
do not consider LULCC, which might be the reason why we overestimate the size
of our carbon pools. Other studies such as suggest a
range of 800–1300 PgC for the global terrestrial biomass. The large range
is a result of inconsistent definitions of forest, uncertain estimates of
forest area, paucity of ground measurements and the lack of reliable
mechanisms for upscaling ground measurements to larger areas
.
A large number of observational-based estimates for global SOC exists with
most studies reporting a global estimate of about 1500 PgC
. SOC pools simulated by ACCESS-ESM1 are somewhat
smaller with 1050 PgC for the PresLAI scenario and about 1200 PgC for the
ProgLAI scenario. However, these numbers agree well with the best estimate of
1260 PgC derived from the HWSD and considering the large
range of 510–3040 PgC of global SOC simulated by CMIP5 models
this is an encouraging result.
The HWSD also provides a spatial distribution of the SOC density, which is
shown in Fig. along with the results from
ACCESS-ESM1. In general there is good agreement between the two ACCESS-ESM1
scenarios, showing a similar pattern, but with a slightly larger density in
the Northern Hemisphere boreal region for the ProgLAI run. The agreement
between the HWSD and ACCESS-ESM1 is also generally good. However, the HWSD
suggest localised hot spots of high SOC density in North America and Siberia,
which are not covered by ACCESS-ESM1. We also underestimate SOC in the
tropics especially in the maritime continent region. On the other hand, both
ACCESS-ESM1 scenarios suggest a high SOC density in the north Asian region,
which is not apparent in the HWSD.
In addition to other environmental constraints such as water, light and
temperature, carbon storage by terrestrial ecosystems may also be limited by
nutrients, predominantly nitrogen and phosphorus
. However, few estimates are available of
total nitrogen and phosphorus pool sizes and their global spatial
distribution is even more uncertain.
Simulated nitrogen pool sizes are shown in Table , and
there is only a small difference between the two ACCESS-ESM1 scenarios. Our
estimate for the nitrogen in the terrestrial biomass is about 6.5 PgN.
Estimates based on field data reconstructions range from about 3.5 PgN
to 10 PgN , which places the
ACCESS-ESM1 results right in the middle of that range. Soil organic nitrogen
pools are simulated to be about 85 PgN for both ACCESS-ESM1 scenarios, which
is slightly low if compared with estimates based on field data (95 PgC
to 140 PgC ).
The terrestrial phosphorus cycle at present day is even less constrained than
the nitrogen cycle and modelling and empirical estimates vary greatly.
ACCESS-ESM1 results suggest a total of 0.35 PgP in the terrestrial
biosphere, which is lower than the estimated range of 0.5–1 PgP by
. Organic soil phosphorus pool sizes differ to some extent
between the two ACCESS-ESM1 scenarios. The PresLAI model run simulates a pool
size of about 10 PgP and the ProgLAI model run gives a pool size of about
12 PgP (see Table ). Other estimates range from about
5 PgP to about 200 PgP with the upper end being assessed as unrealistic
.
Ocean carbon
Surface field assessment
Figure shows the Taylor diagram comparing the mean surface
alkalinity, DIC, temperature and salinity fields. The ACCESS-ESM1 surface
fields are 20-year averages (1986–2005), assessed against observations.
Overlain on this plot are median values from CMIP5. The individual CMIP5
models are listed in Table .
For all variables considered, ACCESS-ESM1 simulations show good spatial
correlations with the observations of better than 0.7. SST shows the highest
correlation (R>0.98) with the observations, demonstrates a similar
magnitude of variability with only a small positive bias. This is very
similar to the response of CMIP5 median that shows a similar negative bias.
ACCESS-ESM1 sea surface salinity (SSS) shows a reasonable correlation with
observations, of similar magnitude to CMIP5 median (about 0.82). However, the
magnitude of the spatial variability is underestimated and there is a bias of
similar magnitude to the CMIP5 median value. ACCESS-ESM1 has known large
regional biases in surface salinity Fig. 16 and these
biases will in turn also impact the simulated alkalinity. Biases in SSS are
not surprising given the challenges with capturing well the hydrological
cycle in ESMs .
Taylor diagram assessing the response of the ACCESS-ESM1 simulations
(circles), and the median of CMIP5 models (diamonds) with observations. The
numbers correspond to (1) alkalinity, (2) DIC, (3) SST and (4) (sea surface)
salinity. For explanation of how to interpret the diagram please see the
text.
Taylor diagram assessing the alkalinity (a) and DIC
(b) of the ACCESS-ESM1 simulation (circle), the median of CMIP5
models (diamond) and the individual members of the CMIP5 ensemble (crosses)
with observations.
The seasonal cycle of NPP anomalies (PgCmonth-1) from
ACCESS-ESM1 in red and SeaWIFS in black calculated
over the period 1998–2005. Overlain on this plot is the CMIP5 the median
(solid green line) and the range 10th and 90th percentiles (shaded).
The integrated sea–air CO2 fluxes over the period
1986–2005 from (a) ACCESS-ESM1 and (b)
.
As anticipated alkalinity shows the poorest correlation with the observations
of all the variables at 0.72. While this is clearly less than the median
value from CMIP5, we note that with for all the CMIP5 median values presented
here, alkalinity also shows the poorest correlation. Encouragingly, the bias
in alkalinity is closer to the observations, and while the variability is
also overestimated it is consistent with CMIP5 values. While some of these
biases are clearly attributable to salinity, to improve alkalinity in
ACCESS-ESM1 will also require further tuning of the export of calcium
carbonate from the upper ocean. For DIC, ACCESS-ESM1 shows a similar
correlation with observations (Fig. ) as the CMIP5 median, but
overestimates the magnitude of the variability when compared with CMIP5 and
observations. The underestimation of the mean value, can be attributed to the
negative alkalinity bias reducing the surface DIC concentration that would be
in equilibrium with the atmosphere.
While assessing the simulated values with the median CMIP5 values provides
valuable insights, it does not allow us to assess the skill of our model with
individual CMIP5 models. To do this the simulated surface DIC and alkalinity
values are compared with individual CMIP5 models (Fig. ). For
alkalinity (Fig. a), the correlation between ACCESS-ESM1 slightly
underestimates correlation returned by the CMIP5 models, but shows a similar,
and in some cases better, magnitude of spatial variability. At the same time
the bias in surface alkalinity it is still within the range of the CMIP5
models, and many cases lower than individual CMIP5 models, but of opposite
sign overestimate alkalinity. For DIC, we see that our simulation sits in the
spread of the CMIP5 correlation and magnitude (Fig. b). Consistent
with alkalinity simulations, we see negative DIC biases and the ACCESS-ESM1 is
not a significant outlier in terms of its magnitude. Overall, our simulation
has comparable skill to the existing CMIP5 models.
Net primary production
To assess the seasonal anomaly of ocean NPP, calculated as the anomaly of
vertically integrated primary productivity through the water column, the
global ocean is broken down into five regions, following .
Figure shows the NPP seasonal anomaly from ACCESS-ESM1, CMIP5
models and SeaWIFS over the (SeaWIFS) observational period 1998–2005. At the
global ocean scale, seasonally we see that the magnitude of NPP from
ACCESS-ESM1 is less than the amplitude of CMIP5 and SeaWIFS, with poor
phasing. This likely reflects the biases in ACCESS-ESM1 toward lower
latitudes, reflecting excess nutrient supply, and utilisation, to the upper
oligotrophic ocean associated with deeper than observed mixed
layers. In the northern and southern subtropical gyres ACCESS-ESM1
(18–49∘ N and 19–44∘ S respectively) appears to
overestimate the amplitude of the observed seasonal cycle when compared with
SeaWIFS. Again this overestimate of NPP is associated with deeper than
observed mixed layers which increase nutrient supply to the oligotrophic
upper ocean. The phase of the NPP in these regions, where agreement between
observations and CMIP5 is very good, is delayed by about 3 months. This
delay may also be explained by a combination of higher (than observed)
concentrations of nutrients and slower than expected biological productions
associated with cool biases, particularly in the Atlantic Ocean allowing the
bloom to occur later.
In the high-latitude Northern Hemisphere, the magnitude of the seasonal cycle
of NPP is not well captured in ACCESS-ESM1. While CMIP5 appears also to
underestimate the magnitude of the seasonal cycle, ACCESS-ESM1 is lower
again. In contrast, in the Southern Ocean the amplitude of the seasonal cycle
of NPP in ACCESS-ESM1 shows good agreement with observations. However, in the
high-latitude oceans the phase of NPP is delayed by about 2 months. This
delay may be attributed to the too shallow mixed layers that exist in these
regions, which means that it is only when mixed layers start to deepen that
biological productivity can start to occur. As a result the remaining growing
season is shorter (than observed) leading to a reduced total productivity.
This may in part explain why the total NPP Northern Hemisphere is much less
than observed.
Interestingly, in the tropical ocean we see very good agreement in the
amplitude of the seasonal cycle with CMIP5 and SeaWIFS. We note, however,
that comparing the phase of the seasonal cycle from ESMs (ACCESS-ESM1 and
CMIP5) with SeaWIFS is not very meaningful in this region, as they all
simulate their own ENSO cycle with their own timing. Therefore, any
comparison over a 20-year period between models has the potential to be
biased by the number of El Niño or La Niña events.
Sea–air CO2 fluxes
Figure shows that, in the period 1986–2005, ACCESS-ESM1 is in
good agreement with the spatial pattern and the magnitude of sea–air
CO2 fluxes of , hereafter referred to as W13. In
the Southern Ocean (44–90∘ S), which is an important net sink of
carbon, ACCESS-ESM1 (-0.77 PgCyr-1) captures a larger annual-mean
uptake than the sea–air CO2 flux of W13, which only estimated an
uptake of -0.18 PgCyr-1. In the southern subtropical gyres
(44–18∘ S) ACCESS-ESM1 (-0.39 PgCyr-1) captures, but
overestimates, the observed sea–air flux of W13
(-0.23 PgCyr-1). In contrast in the Northern Hemisphere
ACCESS-ESM1 underestimates the uptake at -0.36 and
-0.19 PgCyr-1 in the subtropical, and (sub)polar regions
respectively, while W13 estimated the uptake at -0.69 and
-0.54 PgCyr-1 over the same regions. The uptake in the
tropical ocean is well captured, showing very good agreement between
ACESS-ESM1 and W13, which estimate an uptake of -0.56 and
-0.57 PgCyr-1. Spatially the inter-annual variability in
sea–air CO2 flux is presented in a companion paper .
The seasonal cycle (1986–2005) of sea–air CO2 flux
anomalies (PgCmonth-1) from ACCESS-ESM1 (red line) and
observations (; black line). Overlain is the CMIP5
median (solid green line) and the range as the 10th and 90th percentiles
(shaded).
Column inventory of anthropogenic carbon in the ocean
(molCm-2) from (a) ACCESS-ESM1 and from (b)
GLODAP ( for 1994.
The anomaly of the seasonal cycle of the sea–air CO2 fluxes was
assessed against observations of W13 and CMIP5, shown in Fig.
for the period 1986–2005. Here, we see that ACCESS-ESM1 has a larger global
amplitude of sea–air CO2 fluxes than observed (W13) and simulated,
but close to the upper value of the range from CMIP5 models. We also see that
globally the phase of sea–air CO2 fluxes is not well captured in
ACCESS-ESM1, lying outside the range of the CMIP5 models. To better
understand why there are differences between ACCESS-ESM1, CMIP5 and W13 we
separate the response of sea–air CO2 into the same regions as for
NPP, again following .
ACCESS-ESM1 appears to capture well the phase of sea–air CO2 fluxes
in the subtropical gyres. In the northern subtropical gyre in particular, we
see that the amplitude and phase of the seasonal cycle in ACCESS-ESM1 shows
very good agreement with W13, in contrast with other ESMs (CMIP5). In the
southern subtropical gyres, while the ACCESS-ESM1 appears to overestimate the
amplitude relative to the observations, we see very good agreement with CMIP5
models. As anticipated the tropical ocean shows very little seasonality,
nevertheless we do see good agreement with CMIP5 models. However, the
comparison of ACCESS-ESM1 against observations (while shown) is not very
meaningful as W13 is based on values of oceanic pCO2 from
, which does not include El Niño years.
The largest differences are seen in the representation of sea–air CO2
fluxes in the high-latitude ocean. In the high-latitude Northern Hemisphere,
we see that the magnitude is larger than either CMIP5 or W13 and shows poor
phasing. While the magnitude of the seasonal cycle in the Southern Ocean lies
within the upper range of CMIP5 again poor phasing is seen. That the seasonal
cycle is out of phase suggests that during the summer the solubility response
likely dominates over the NPP response, leading to an outgassing in the
summer and uptake in the winter, as discussed in .
Consequently, we see that the poor global phasing in global sea–air
CO2 fluxes is likely due to the solubility dominated response of the
high latitudes during the summer.
Anthropogenic inventory
The global inventory of anthropogenic carbon from ACCESS-ESM1 is compared
with the uptake from GLODAP for the year 1994 in
Fig. . Here we see that the spatial pattern of the column
inventory of anthropogenic carbon is very well reproduced, with the large
storage occurring in the North Atlantic and large uptake in the Southern
Ocean. The inventory for the period 1850–1994 in ACCESS-ESM1 is 132 PgC,
which is close to the estimated value from GLODAP of 118±19 PgC
over the same domain. This suggests that despite a
somewhat limited representation of the seasonal cycle of sea–air CO2
fluxes in key regions of anthropogenic uptake, such as the Southern Ocean,
the ACCESS-ESM1 is doing a very good job, spatially and temporally, of
capturing and storing anthropogenic carbon. If the entire domain (including
the Arctic Ocean) is integrated, the anthropogenic uptake is 143 PgC over
the same period.
Mean seasonal cycle of atmospheric CO2 for the period
1986–2005 from land carbon fluxes (dashed lines) and both land and ocean
carbon fluxes (solid line). The prescribed LAI case is shown in blue, the
prognostic LAI case in red and observations based on flask data from
GLOBALVIEW in black for (a) Alert (82.45∘ N,
62.52∘ W), (b) Mace Head (53.33∘ N, 9.90∘ W),
(c) Mauna Loa (19.53∘ N, 155.58∘ W) and (d)
the South Pole (89.98∘ S, 24.80∘ W).
Atmospheric CO2
The land and ocean carbon fluxes have been put into two atmospheric tracers
as described in Sect. 2.4. These tracers have no impact on
the model simulation but allow for the atmospheric CO2 distribution to be
assessed. A reasonable simulation of known features of atmospheric
CO2 can increase our confidence in the simulated carbon fluxes. For
example the seasonal cycle of atmospheric CO2 is strongly driven by
the seasonality in land carbon fluxes. Therefore, our simulated seasonality
can be realistically compared to present-day atmospheric CO2
observations.
The seasonal cycle of atmospheric CO2 is shown for four locations at
different latitudes (Fig. , note the different vertical
scale in the upper and lower panels). Seasonal cycles from the PresLAI and
ProgLAI cases are calculated as the mean over the last 20 years of the
historical period (1986–2005) with the annual mean removed from each year.
The seasonality is plotted for the contribution from the land carbon fluxes
only and for both the land and ocean carbon fluxes combined. The model output
was taken from the nearest grid point to each location with the exception of
Mace Head, where the model was sampled further west to better approximate the
observations, which are selected for clean-air (ocean) conditions.
As observed, the amplitude of the seasonal cycle decreases from north to
south. At Alert (82∘ N, Fig. a) both model
simulations overestimate the seasonal amplitude by up to 6 ppm with
the growing season starting earlier than currently observed. The ocean carbon
fluxes contribute little to seasonality at this latitude. At Mace Head
(53∘ N, Fig. b) the simulated seasonal cycle is
comparable to that observed with only a small difference in the seasonal
amplitude (smaller than 2 ppm), while at Mauna Loa (20∘ N,
Fig. c) the ProgLAI case better represents the observed
seasonality than the PresLAI case.
Seasonal cycles in the Southern Hemisphere (e.g. South Pole) are more
challenging to simulate correctly as they are made up of roughly equal
contributions from local land fluxes, Northern Hemisphere land fluxes and
ocean fluxes. Figure d shows for the PresLAI case that
the simulated seasonality from the land carbon fluxes is shifted in phase
when the ocean carbon contribution is included but the phase shift is away
from the observed seasonality. This phase shift is not apparent for the case
with ProgLAI.
Conclusions
The evaluation of ACCESS-ESM1 over the historical period is an essential step
before using the model to predict future uptake of carbon by land and oceans.
Here, we performed two different scenarios for the evaluation of the land
carbon cycle: running ACCESS-ESM1 with a prescribed LAI and a prognostic LAI.
Running with a prognostic LAI is our preferred choice, since this includes
the vegetation feedback through the coupling between LAI and the leaf carbon
pool. However, results have shown that we overestimate the amplitude of the
prognostic LAI annual cycle in the Northern and Southern hemispheres and
underestimate it in the tropics. In future versions we need to improve the
performance of the prognostic LAI, particularly for evergreen needle leaf and
C4 grass.
ACCESS-ESM1 shows a strong cooling response to anthropogenic aerosols, which
is offsetting the warming due to increases in greenhouse gases. The aerosol
radiative forcing over the historical period is much stronger than the IPCC
best estimate, but still within the uncertainty range. The impact of the
cooling due to anthropogenic aerosols in ACCESS-ESM1 needs to be quantified
in future work.
The land carbon uptake over the historical period is about 40% larger
for the run with prognostic LAI in comparison to the run with prescribed LAI.
This is mainly due to the stronger response to volcanic eruptions, which
increases GPP in the tropics and reduces plant respiration globally,
therefore increasing NEE.
Globally integrated sea–air CO2 fluxes are well captured and we
reproduce very well the cumulative uptake estimate from the Global Carbon
Project and our anthropogenic uptake agrees very well
with observed GLODAP value of . The spatial distribution of
sea–air CO2 fluxes is also well reproduced by CMIP5 models and
observations. At the same time global ocean NPP also shows good agreement
with observations and lies well within the range of CMIP5 models. However,
seasonal biases do exist in sea–air CO2 fluxes and NPP, potentially
related to biases in MLD and surface temperature that are
present in ACCESS-ESM1, and will need to be addressed in later versions of
ACCESS-ESM1.
Simulated carbon pool sizes are generally within the range of estimates
provided in the literature. Simulated soil organic carbon has been compared
against the Harmonized World Soil Database, finding very good agreement in
the spatial distribution and the total size. Nitrogen and phosphorus
limitation were active in our simulations and pool sizes seem reasonable if
compared with other estimates. However, nitrogen and phosphorus cycles are
poorly constrained and only a few global estimates exist with large
uncertainties.
ACCESS-ESM1 has the capability of putting land and ocean carbon fluxes into
tracers, which provides a way of assessing simulated atmospheric CO2
concentrations. The simulated seasonal cycle is close to the observed, but we
overestimate the amplitude in the high northern latitude by up to
6 ppm and we also notice small phase shifts.
Overall, land and ocean carbon modules provide realistic simulations of land
and ocean carbon exchange, suggesting that ACCESS-ESM1 is a valuable tool to
explore the change in land and oceanic uptake in the future.