Earth System Models (ESMs) that incorporate carbon-climate feedbacks represent the present state of the art in climate modelling. Here, we describe the Australian Community Climate and Earth System Simulator (ACCESS)-ESM1 that combines existing ocean and land carbon models into the physical climate model to simulate exchanges of carbon between the land, atmosphere and ocean. The land carbon model can optionally include both nitrogen and phosphorous limitation on the land carbon uptake. The ocean carbon model simulates the evolution of nitrate, oxygen, dissolved inorganic carbon, alkalinity and iron with one class of phytoplankton and zooplankton. From two multi-centennial simulations of the pre-industrial period with different land carbon model configurations, we evaluate the equilibration of the carbon cycle and present the spatial and temporal variability in key carbon exchanges. For the land carbon cycle, leaf area index is simulated reasonably, and seasonal carbon exchange is well represented. Interannual variations of land carbon exchange are relatively large, driven by variability in precipitation and temperature. We find that the response of the ocean carbon cycle shows reasonable agreement with observations and very good agreement with existing Coupled Model Intercomparison Project (CMIP5) models. While our model over estimates surface nitrate values, the primary productivity agrees well with observations. Our analysis highlights some deficiencies inherent in the carbon models and where the carbon simulation is negatively impacted by known biases in the underlying physical model. We conclude the study with a brief discussion of key developments required to further improve the realism of our model simulation.
Over recent decades many climate models have evolved into earth system
models (ESMs), a term used to identify models that simulate
biogeochemical cycles and their interaction with human and climate
systems. Of principal concern is the carbon cycle. Anthropogenic
emissions of carbon lead to increased concentrations of atmospheric
carbon dioxide (
The Coupled Model Intercomparison Project (CMIP5)
The Australian Community Climate and Earth System Simulator, ACCESS,
has been developed over recent years to meet both the numerical
weather prediction
ACCESS-ESM1 comprises the ACCESS1.4 physical climate model
(Sect.
As described in
The physical model to which we are adding the carbon cycle is derived
from ACCESS1.3 and designated ACCESS1.4. ACCESS1.4 addresses a number
of issues that were identified during the analysis of the ACCESS1.3
CMIP5 simulations and also includes an updated version of CABLE
(CABLE2). Changes made to CABLE are discussed in
Sect.
ACCESS1.3 used atmospheric physics settings similar to the Met Office
Global Atmosphere (GA) 1.0 configuration
Analysis of ACCESS1.3 simulations showed almost no dust in the
atmosphere
In addition to the change in dust, the ACCESS1.3 control simulation did not include background stratospheric volcanic forcing but this has been included in ACCESS1.4 simulations. Preliminary tests with the dust and vocanic forcing changes reduced the globally averaged surface air temperature relative to ACCESS1.3. Since an aim of ACCESS1.4 was not to change global-scale climate characteristics relative to ACCESS1.3, one of the parameters in the cloud scheme (FW_STD associated with the standard deviation of cloud water content) was increased from 0.700 in ACCESS1.3 to 0.725 in ACCESS1.4. This resulted in a globally averaged surface air temperature in ACCESS1.4 that was similar to that obtained for ACCESS1.3. ACCESS1.4 also corrects a bug which zeroed the downward short-wave radiation over coastal sea-ice points for non-radiation timesteps. This reduced excess ice accumulation in ACCESS1.3 in some coastal regions such as the Canadian Archipelagos.
While there are no changes in the ocean model version between
ACCESS1.3 and ACCESS1.4, there have been two changes in the
configuration or parameter values. Firstly for ACCESS1.4, the
background vertical diffusivity outside 20
ACCESS1.3 used the OASIS3.2-5 coupler
CABLE is a land surface model that simulates the fluxes of
momentum, heat, water and carbon across the land-atmosphere
interface. CABLE operates both in standalone mode (forced with
prescribed meteorology) and coupled to atmospheric models (at
least five different models to date, both global and regional).
The history and scientific core of CABLE version 1 is most fully
described in
ACCESS1.4 and ACCESS-ESM1 use CABLE2.2.3
(Fig.
In ACCESS, CABLE is run for one or more tiles in each grid-cell
with a non-zero land fraction. Each tile represents a different
vegetated or non-vegetated surface type with a number of CABLE
input parameters being surface type dependent
(Sect.
The flux of carbon from the land to the atmosphere has two
components, net ecosystem exchange (NEE) and fluxes due to
disturbance (e.g. fire) and land-use change. Currently CABLE
simulates the former, as the difference between respiration and
photosynthesis, but not the latter. Thus
GPP and leaf maintenance respiration are calculated every time step
using a two-leaf (sunlit and shaded) canopy scheme
Daily mean GPP and leaf respiration are passed into the
biogeochemical module which is run once per day to calculate the
remaining respiration fluxes and the carbon flow between pools.
The fractions of GPP allocated to each
vegetation pool are vegetation dependent parameters which, for
non-evergreen vegetation types, are also dependent on leaf
phenology phase
Maintenance respiration of woody tissue and roots and growth
respiration are calculated as a function of mean daily air
temperature and tissue nitrogen amount. Default carbon to nitrogen
and nitrogen to phosphorus ratios are used when nitrogen and/or
phosphorus are not simulated. Growth respiration is calculated
daily as a proportion of the difference between daily GPP and plant
maintenance respiration, with the proportion being a function of
leaf nitrogen to phosphorus ratio
Since plant and soil respiration rates are only calculated daily, CABLE in ACCESS-ESM1 is not expected to realistically simulate the diurnal cycle of the net land carbon flux to the atmosphere, and we restrict our analysis to monthly or longer timescales.
Carbon should be conserved across the land carbon system, that is
the net flux to the atmosphere over a given time period should
equal the change in the total carbon across all carbon pools over
that same period. A carbon conservation check is presented in
Sect.
CABLE with CASA-CNP has been used in a number of offline
applications, where meteorological forcing is prescribed,
The Whole Ocean Model of Biogeochemistry And Trophic-dynamics
(WOMBAT) model is based on a NPZD (Nutrient, Phytoplankton,
Zooplankton and Detritus) model with the additions of bio-available
iron limitation (
In this model we include two DIC tracers: natural and anthropogenic
DIC. These two DIC tracers only differ in the atmospheric
ACCESS-ESM1, mostly through capability inherited from the Met
Office Unified Model, has the option of running with or without
interactive
When ACCESS-ESM1 is run without interactive
The atmospheric transport of
The ACCESS-ESM1 atmosphere is run with a horizontal resolution of
1.875
As noted above, CABLE can simulate land carbon fluxes with or
without nutrient limitation. Here we have chosen to run CABLE in
the “CNP” configuration, based on results from some low
resolution ESM studies.
For most of the work described here, two sets of simulations have
been performed. In the first set, leaf area index is prescribed and
there should be no interaction between the carbon cycle and the
climate simulation (given that atmospheric
Most of the input files and parameter settings (Supplement) for
the biophysical component of CABLE were as described in
Differences between the model configuration here and
Additional input files are required for the biogeochemistry module
of CABLE and these are based on
To simulate nitrogen and phosphorus requires nitrogen deposition
and fixation, phosphorus from weathering and from dust and soil
order, to distinguish soils of different mineralogy and age. These
are all taken from
The initial conditions for phosphate (
There was no formal spin-up of the carbon cycle before the ACCESS-ESM1 pre-industrial control run was started. The land carbon pools were initialised at values taken from repeated test simulations using the prognostic LAI configuration. The ocean BGC initial fields come from the observed climatology as described in the previous section. Offline land simulations and ocean-only simulations were explored to aid in the spin-up process but neither produced a satisfactory result at the time the pre-industrial run was started. This partly reflected the significant and evolving change of the mean climatology of the land, ocean and atmosphere from the present-day state.
In this section results from two ACCESS-ESM1 pre-industrial control
simulations will be characterised and compared. Each simulation
presented here used prescribed (rather than interactive)
atmospheric
A brief analysis of the simulated climate is presented first
(Sect.
Relative to the range of CMIP5 models, the two ACCESS submissions,
ACCESS1.0 and ACCESS1.3 produced similar results when various
modelled atmospheric climate variables were compared against
observations
For surface air temperature, the prognostic LAI case results in
globally warmer temperatures (
To provide a perspective on how the ocean dynamics changes between
ACCESS1.3 and ACCESS1.4 we compare the global meridional
overturning streamfunction and the annual maximum mixed layer
averaged over the last 100
Any climate model produces biases in its climate simulation when
compared with observations. Some of these biases may also have
implications for the simulation of the carbon cycle. Here we note
two biases that impact on different components of the carbon
cycle. Firstly,
The conservation of land carbon has been checked across a sample
100
If we choose
Tiles with poor carbon conservation are characterised by zero or
very low leaf carbon and possibly other highly depleted carbon
pools. The magnitude of the carbon imbalance is well correlated
(greater than 0.9) with a count of the number of months with zero
leaf carbon across the 100
The temporal evolution of the global land carbon fluxes over the
1000
Figure
The slightly positive NEE flux to the atmosphere is balanced by
a decrease in the total carbon across all pools
(Fig.
The behaviour of the nitrogen pools (not shown) is broadly similar to the carbon pools with nitrogen loss from the passive soil pool, again largely from the evergreen broadleaf vegetation type. This loss is offset, to a greater extent than for carbon, by increases in nitrogen in the slow soil pool, primarily for the tundra vegetation type. The trend in pools is a little different for phosphorus with both the passive and slow soil pools growing, while the inorganic phosphorus pools are declining. As for nitrogen the slow soil pool change is dominated by the tundra vegetation type but the other pool changes are split more evenly across a range of vegetation types.
The zonal mean GPP, plant and soil respiration over the last
500
The difference in GPP can be understood when the prognostic LAI is
compared to the LAI values used in the prescribed (PresLAI) case
(Fig.
Land carbon fluxes are highly seasonal and this is captured by the
model; Fig.
Including prognostic LAI in the simulation changes the interannual
variability (IAV) of the land carbon fluxes. For global fluxes
(Table
The impact of climate variability on NEE is seen in
Fig.
WOMBAT conserves the biogeochemical tracers in the ocean, which
means the rate of change in the total carbon in the ocean equals
the net sea–air flux, noting that the sea–air flux is negative for
Within WOMBAT, if particulate organic matter and calcium carbonate
are not remineralized before reaching the seafloor they can
accumulate in the sediments. Our simulations show that the carbon
in the sediments are stable and small (Fig.
To assess our ocean carbon cycle simulation and CMIP5 simulations
Encouragingly all variables from the ACCESS-ESM1 simulation show
correlations with the observations of close to 0.6 or better. SST
shows a very high correlation (
DIC in ACCESS-ESM1 shows a good correlation with observations
(Fig.
While assessing the simulated values with the median CMIP5 values
provides valuable insight, it does not allow us to assess the skill
of our model when compared with individual CMIP5 models. To this
end the simulated state variables of the carbon system, DIC and ALK
are compared with individual CMIP5 models (Fig.
The sea–air carbon flux is shown in Fig.
Documentation of ACCESS-ESM1 and its performance under
pre-industrial, prescribed atmospheric
Analysis of the pre-industrial simulation has highlighted some issues with the ACCESS-ESM1 carbon models and how biases in the physical model simulation can contribute to a poor simulation of carbon fluxes. For land carbon, a high priority is to fix the inability of CABLE to conserve carbon in situations where moisture is insufficient to maintain vegetation and to confirm whether land carbon fluxes are too sensitive to climate (particularly rainfall) variability. Development priorities for CABLE in future ACCESS-ESM versions are implementation of land use change, the ability for phenology to respond to climate and improved nutrient forcing (e.g. temporally varying input fluxes).
In the ocean we see reasonable agreement with observations, and
results that fall within the range of existing CMIP5 models for DIC
and alkalinity. The spatial pattern of pre-industrial sea–air
carbon fluxes shows very good agreement with published studies,
while primary productivity is close to the observed
value. Nevertheless there are outstanding issues to be addressed in
the ocean: (a) reducing salinity biases would improve the simulated
values of alkalinity and DIC, bringing these closer to the
observations; and (b) reducing the excess of surface nitrate,
potentially through modifying the particulate organic carbon
export. Furthermore we see a recognised need to add additional
complexity, in terms of phytoplankton and zooplankton classes, to
capture the potential impacts related to projected changes in the
marine environment such as ocean acidification
It is clear from our simulations that our model has yet to fully
reach quasi-steady state, despite more than 1000
At present the next physical model version of ACCESS (ACCESS-CM2) is currently being developed in preparation for CMIP6. The land and ocean carbon cycles presented here will form the basis for ACCESS-ESM2.
Code availability varies for different components of
ACCESS-ESM1. The UM is licensed by the UK Met Office and is not
freely available. CABLE2 is available from
This research is supported by the Australian Government Department of
the Environment, the Bureau of Meteorology and CSIRO through the
Australian Climate Change Science Programme. The research was
undertaken on the NCI National Facility in Canberra, Australia, which
is supported by the Australian Commonwealth Government. The authors
wish to acknowledge use of the Ferret program for some of the analysis
and graphics in this paper. Ferret is a product of NOAA's Pacific Marine Environmental
Laboratory. (Information is available at
Model Parameters of the BGC model were set to the values optimised in the 1-D model of the Southern Ocean
Standard deviation of annual global carbon flux for years 901–1000 in
Schematic showing the different component models of ACCESS-ESM1 and the ACCESS versions on which it is dependent.
Root mean square difference (RMSD) between atmospheric variables simulated by
the model versions listed in the key and those from the ACCESS1.3
pre-industrial simulation normalised by the RMSD between ACCESS1.0 and
ACCESS1.3. The variables are precipitation (pr), surface air temperature
(tas), sea level pressure (psl), top of atmosphere long-wave radiation
(rlut), top of atmosphere reflected short-wave radiation (rsut), air
temperature (ta), zonal (ua) and meridional wind (va) at 850 and
200
Global Meridional Overturning Streamfunction (Sv) from 100
Maximum mixed layer (m) from 100
25
100
Zonal mean year 501–1000 carbon flux
Monthly mean NEE in
Standard deviation of annual NEE
Correlation between annual NEE and
Global
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) Nitrate, (2) Alkalinity, (3) DIC, (4) SST, and (5) (sea surface) Salinity. For explanation please see the text.
Taylor diagram assessing the DIC
The mean sea–air flux of carbon dioxide for the years 901–1000