This paper provides a comprehensive description of OSCAR v2.2, a simple Earth system model. The general philosophy of development is first explained, followed by a complete description of the model's drivers and various modules. All components of the Earth system necessary to simulate future climate change are represented in the model: the oceanic and terrestrial carbon cycles – including a book-keeping module to endogenously estimate land-use change emissions – so as to simulate the change in atmospheric carbon dioxide; the tropospheric chemistry and the natural wetlands, to simulate that of methane; the stratospheric chemistry, for nitrous oxide; 37 halogenated compounds; changing tropospheric and stratospheric ozone; the direct and indirect effects of aerosols; changes in surface albedo caused by black carbon deposition on snow and land-cover change; and the global and regional response of climate – in terms of temperature and precipitation – to all these climate forcers. Following the probabilistic framework of the model, an ensemble of simulations is made over the historical period (1750–2010). We show that the model performs well in reproducing observed past changes in the Earth system such as increased atmospheric concentration of greenhouse gases or increased global mean surface temperature.
Simple biogeochemistry–climate models, also qualified as compact or
reduced-form models, are widely used in the climate change research
community. These models share several features. First, they are not spatially
resolved and as such they can be referred to as box models, although the
number of boxes – and therefore of state variables – may vary greatly: from
a couple to several hundred. This limited number of state variables make
them relatively intelligible, when compared to complex three-dimensional
models. Second, the time step of analysis and of numerical solving is about
1 year, which implies they usually cannot include representations of
seasonal processes. One consequence of these two features is their very high
computing efficiency: one simulation typically takes about 1 min on a
laptop. Compact models can therefore be used in a variety of setups, such as the following for
instance: to translate a large number of pathways of greenhouse gases
emissions into projected climate change
Here, we present an important update of a simple Earth system model that has already been used for some time. The model is named OSCAR, and this paper provides a comprehensive description of version 2.2. OSCAR can be described as a non-linear box model of which the number of boxes, however, is fairly large. It is not spatially resolved (i.e., it is not gridded) but key processes such as land-use change or aerosol physico-chemistry are regionalized to account for the disparity in such processes that is observed in the real world. OSCAR does not endogenously simulate intra- or inter-annual variability. Consequently, although the time step of its inputs and outputs is 1 year, the main purpose of the model is to simulate trends in the Earth system change, and not year-to-year variations. OSCAR is also a parametric model in which a relatively large number of parameters are almost all calibrated on complex models. We call this approach meta-modeling: each module of OSCAR is designed to emulate the behavior of other more specialized models (e.g., global climate models, dynamical vegetation models, or chemistry-transport models). For most modules, we have access to several sets of parameters (one per complex model used to calibrate) and, rather than taking the average or arbitrarily choosing one, we adopt a probabilistic approach in which a given simulation with OSCAR is repeated many times with different sets of parameters picked at random.
This paper is firstly a thorough description of OSCAR. Readers who are more
interested in the model's applications should know that only a simulation
covering the historical period is made here, for diagnostic and (partial)
validation purposes. More concrete applications of the model or of its older
versions can be found elsewhere in the literature
Section 2 provides the details of the mathematical formulation of the model, and it describes how the parameters are calibrated or derived. The first subsection provides general information about the model. The second subsection describes the forcings of the model: anthropogenic emissions of greenhouse gases, ozone precursors, aerosols and aerosols precursors, land-use and land-cover change, and some additional anthropogenic and natural radiative forcings. The next subsections describe the model's various modules dedicated to the prediction of carbon dioxide, methane, nitrous oxide, halogenated compounds, ozone, aerosols, surface albedo, and the climate response. The last subsection describes the numerical solving method. Section 3 then provides the first results from OSCAR v2.2 in the case of a simulation over the historical period from 1750 to 2010, as well as a brief discussion of these results. Section 4 provides some preliminary conclusions regarding the model, its performance and interest, and future potential development.
Since version 2.0 (see Appendix A for an overview of the model's history),
the development of OSCAR has been driven by three principles. First, we aim to
embed in OSCAR as many components and processes of the actual Earth system as
possible, with the overall idea of favoring the amount of processes,
interactions, and feedbacks implemented over the modeling complexity of each
of these elements. A diagram of the model's simplified causal structure is
shown in Fig. 1. Second, OSCAR is built as a meta-model capable of
emulating the sensitivities of models of higher resolution or superior
complexity. To do so, in order to calibrate those sensitivities, we use
outputs from complex models that participated in intercomparison projects
whenever possible. Consequently, OSCAR is designed to be used in a
probabilistic framework. The last table of the paper summarizes the various
options available for this probabilistic setup. Third, the model is developed
as a dynamic model of the Earth system with an assumed equilibrium
corresponding to the preindustrial era. The reason for this is the original
purpose of the model, to perform studies attributing the anthropogenic
component of climate change
Simplified causal chain of OSCAR v2.2. Each node of the graph
corresponds to a module described in the section whose number is shown below
the node's name. Colored lines show the forcings of the model, black lines
show the natural cause-effect chain, and dashed lines show the climate
feedbacks. “Halo” groups all the halogenated compounds, “Ocean” is the ocean
carbon cycle, “Land” is the land carbon cycle, “Albedo” groups the surface
albedo effects, “
This last point is also the reason why all the following equations are
expressed as a difference to our so-called preindustrial equilibrium. In the
model, all variables
Let us now introduce some of the main notations that are used throughout the
description sections. The variables of the model are written with Roman
letters, whereas its parameters are with Greek letters. Among the variables,
some letters are consistently dedicated to a specific kind of variable:
Anthropogenic emissions of various active gases are the main drivers of
climate change, and thus of OSCAR. In this version of the model, the
anthropogenically emitted greenhouse gases directly prescribed as emissions over the
1750–2010 period are carbon dioxide from fossil-fuel
burning and cement production (
Because most of these anthropogenic drivers are actually poorly known,
especially when going backward in time, we use various established
inventories in order to introduce variation in our model's results, and thus
explore the uncertainty in climate change reconstruction and projection. The
historical emissions of fossil-based CO
Similarly to what is done by
Finally, because of our assumption of a preindustrial equilibrium occurring before 1750, we have to offset the full time series of anthropogenic emissions obtained thus far by the level of emissions of 1750, thus assuming that everything before that point in time is part of the natural cycle – or at least negligible as an anthropogenic perturbation of this natural cycle. The obtained time series of anthropogenic emissions for all species are shown in Fig. 2, except for the halogenated compounds which are shown in Fig. S1 in the Supplement.
As OSCAR embeds a book-keeping module to endogenously calculate CO
Time series of the main anthropogenic emissions used as potential inputs of OSCAR (Sect. 2.2.1). Other drivers of the model, i.e., emissions of halogenated compounds and LULCC, are shown in Figs. S1 and S2, respectively.
In this version of OSCAR, only one LULCC dataset is available: the LUH v1.1
dataset
Finally, some remaining known climate forcings are prescribed to OSCAR
directly as radiative forcings. This is the case of one anthropogenic
forcing (aviation contrails and induced cirrus (
The carbon cycle in OSCAR is divided in two components: ocean and land. The
ocean carbon cycle includes the so-called physical pump, i.e., the dissolution
of anthropogenic CO
The ocean carbon-cycle module is based on the mixed-layer impulse response
developed by
Following
The atmospheric conversion factor is calculated following
We use the latter study in a way very close to what is done by
The two parameters related to the mixed-layer depth, i.e.,
Before considering the extensive perturbation of the terrestrial
carbon cycle, driven by land-use and land-cover changes, we first focus on
its intensive perturbation, i.e., the perturbation that changes the areal
properties of the terrestrial ecosystems. This intensive perturbation is
driven by changes in environmental conditions such as atmospheric CO
The terrestrial carbon-cycle module is an upgrade of the previous versions
Contrarily to complex models that simulate separately the gross primary
productivity, our terrestrial carbon cycle starts directly with the net
primary productivity (NPP). The areal net productivity (
The equation system described above by Eqs. (19), (20), and (21) implies
that our preindustrial equilibrium is as follows.
The parameters for the preindustrial fluxes (i.e.,
The parameters for the transient response of NPP and
heterotrophic respiration (i.e.,
The fire-related parameters are similarly calibrated on TRENDY (for
Regarding the TRENDY and CMIP5 data processing, it has to be noted that none of the models provide biome-specific outputs. So we choose to deduce biome-specific data by weighting the biome-aggregated outputs of a model by its biome area fraction map, taken to the power 3. This approach is used to give more importance – in a given region – to the grid cells in which biomes are purer, without taking the risk of having too few of those grid cells if we were to set a threshold of biome area fraction instead. Also, some of the complex models used to calibrate OSCAR are lacking some of the biomes implemented in our model. Thus, we need rules to establish parameters for the lacking biomes on the basis of the available ones. When croplands are not in a model, we assume they have the same biogeochemical properties as grasslands, before any harvest or other human intervention. When pastures are not in a model, we assume their biogeochemical parameters are a mix of those of grasslands and bare soil, at 60 and 40 % respectively. In a configuration of OSCAR in which shrublands are separated from grasslands – which is not the case in this paper – and shrublands are not in a model, we assume they are made at 85 % of grasslands and 15 % of forests.
The preindustrial area extents
Now we consider the extensive perturbation of the terrestrial carbon cycle,
i.e., the one driven by changes in land use and land cover. This perturbation
has a first-order effect that originates from the human-induced disturbance
of a given biome which then transitions from its disturbed state to a new
steady-state. When both extensive – change in biome extent – and intensive
– change in areal properties – perturbations occur at the same time, their
interaction creates a second-order effect, which is also included in the
following equations. Here, we also note that in theory another extensive
perturbation affects the terrestrial ecosystems: the migration of natural
biomes caused by changes in environmental conditions
The book-keeping module used to estimate the carbon fluxes induced by the
land-use drivers is very close to that of the previous version of OSCAR
For the book-keeping itself, we need to define a new series of extensive
state variables for the terrestrial biosphere affected by LULCC (subscript
The initialization of the book-keeping sequence, i.e., the initial disturbed
state of a given triplet
In the case of land-cover change and shifting cultivation, the above-ground
fraction (
Once the initialization round is done, the LUC-disturbed biospheric pools
follow the same carbon cycle as the one described in the previous section for
undisturbed biomes:
As for the harvested wood products, they are oxidized at a varying rate that
depends on the characteristic time of the pool (i.e., on
Functional forms possible for the harvested wood products oxidation
(Sect. 2.3.3). They are shown as the oxidation profile of a unit pool of
wood product (left-hand panel) and as the corresponding normalized yearly
oxidation rate (right-hand panel). The former is denoted by
The
The incremental change in atmospheric CO
In Eq. (33) this version of OSCAR notably ignores the permafrost carbon
that may be emitted under a warming climate
The radiative forcing (RF) induced by the increase in atmospheric CO
This intermediary section is dedicated to two elements which will be needed
hereafter for non-CO
The atmospheric CO
The
In the next sections, we need an estimate of the stratospheric concentration
change of some species. For relatively long-lived species, we assume the
stratospheric concentration change of this species can be approximated by its
change in atmospheric concentration (
We set
In OSCAR, all known atmospheric sinks of methane are included, and particular attention is paid to how the main tropospheric sink varies with anthropogenic and natural external factors. Amongst the natural sources of methane, only the emissions from wetlands are endogenously calculated in the model, implying that all other natural sources – such as freshwaters, termites, or permafrost – are assumed to remain constant.
The oxidation of atmospheric methane follows the same modeling approach as
that of the previous version
The flux of oxidized CH
The four lifetimes of methane are taken as the present-day lifetimes given by
All the chemical sensitivities of the OH sink (i.e.,
Natural wetlands are the largest natural source of methane
First, we estimate the regional change in CH
We calibrate two sets of parameters for wetlands. First, the preindustrial
equilibrium of the wetlands can be calibrated on seven WETCHIMP models
On the basis of the previous sections, the incremental change in atmospheric
CH
The radiative forcing induced by the increase in atmospheric CH
In OSCAR, the stratospheric sink of nitrous oxide is included, and particular attention is paid to how it varies with anthropogenic and natural external factors. However, no other natural processes are endogenous to the model, meaning that no change in natural sources or sinks of nitrous oxide (e.g., ocean, natural soils, biological fixation) is assumed.
The oxidation of nitrous oxide follows the same modeling approach as that of methane, with only one sink in the stratosphere that has a varying lifetime. The law used to make the stratospheric lifetime vary, however, is recent and different from the previous version of the model.
The flux of oxidized N
The preindustrial stratospheric lifetime
Regarding the chemical sensitivity to the age of air, we assume it is not
zero only when the other sensitivities are deduced from the “G2d” model,
therefore following the results of
The incremental change in atmospheric N
Similarly to methane, the radiative forcing induced by the increase in
atmospheric N
OSCAR accounts for many halogenated species. These are grouped into three
categories: 11 hydrofluorocarbons (HFC-23, HFC-32, HFC-125, HFC-134a,
HFC-143a, HFC-152a, HFC-227ea, HFC-236fa, HFC-245fa, HFC-365mfc,
HFC-43-10mee) noted together as
Conceptually, the modeling approach of the halogenated compounds' sinks is
similar to that used for methane. Each of these species
The lifetimes
The incremental change in atmospheric concentration of any species
The radiative forcing induced by the increase in atmospheric concentration of
any of those species
In OSCAR, as in other simple models
In our model, change in tropospheric ozone is driven by atmospheric methane,
emissions of ozone precursors, and global climate change. We use a
formulation close to that of the previous version of OSCAR, which was the
formulation by
The change in global tropospheric ozone burden (
The global chemical sensitivities (i.e.,
The regional weights
Finally, the radiative forcing induced by the change in tropospheric ozone
burden is assumed to be linear:
With the same formalism as for tropospheric ozone, change in stratospheric ozone is driven by available stratospheric chlorine and bromine, stratospheric nitrous oxide, and global climate change. Compared to the previous version that was using only a linear dependency on chlorine and bromine, nitrous oxide and climate change are two new drivers in this module.
The first step to model stratospheric ozone is to estimate its first driver
of change: the stratospheric chlorine and bromine available from the presence
of the ODSs in the stratosphere. Those compounds release their chlorine
and/or bromine atoms at various rates and thus interact differently with
ozone. A proxy variable is thus created to lump together these various
effects, namely the equivalent effective stratospheric chlorine (EESC). The
EESC is calculated following
Regarding the EESC parameterization,
The chemical sensitivity of stratospheric ozone to EESC and that to global
climate change (i.e.,
Finally, the radiative forcing induced by the change in stratospheric ozone
burden is assumed to be linear:
As for ozone, because the aerosols are short-lived, it is assumed that their
global atmospheric burden reaches a steady-state with their respective
drivers of change at each time step of the model. The direct and indirect
radiative effects of five anthropogenic aerosols are considered here, with
the notable caveat that their atmospheric physico-chemistry is only loosely
coupled to the rest of OSCAR: for instance, the oxidation of SO
The direct effect of aerosols refers to the direct radiative forcing caused
by the aerosol–radiation interactions, i.e., without consideration of any
short-term adjustment of the climate system
It must be noted that here we purposefully limit the number of these drivers of change: only two precursors are considered for each aerosol, to avoid over-fitting on data, which does not allow us to clearly separate the effect of each precursor, and we add the global surface temperature, used as a proxy of a changing climate. For the same reason – because of the calibration data – we keep the modeling simple with linear sensitivities. Note also that in this section every lifetime is said to be “apparent”, because it corresponds to a globally averaged chemical sensitivity that has dimensions of time, and which results from several physical and/or chemical processes not explicitly modeled in OSCAR.
In the case of sulfate aerosols, their change in burden (
Finally, the change in burden of secondary organic aerosols
(
For SO
The regional weights
For any of the five aerosols
Under this term, we group the so-called semi-direct and indirect effects –
that is, the rapid adjustments in the atmospheric system induced by
aerosol–radiation interactions and the adjusted aerosol–cloud interactions,
according to the terminology by
For the semi-direct effect, the modeling approach is straightforward.
According to
One possible value for the coefficient used to account for the semi-direct
effect is based on the fifth IPCC report
The derivation of the parameters for the aerosol–cloud interaction is done in
three steps. First, we need the soluble aerosol fractions
Second, we calculate the intensity parameter
Third, because this preliminary value of
Anthropogenic perturbations of the Earth's energy budget through surface albedo change are difficult to model in a simple way, because they are local phenomena with significant seasonal variability. Moreover, they can involve non-radiative processes that are almost impossible to capture with simple models. The two OSCAR modules presented hereafter are first-order models of two surface albedo perturbations: BC deposition on snow, and land-cover change. As such, they are not coupled with one another, nor are they with the climate module.
Ensemble of possible parameterizations of the aerosol–cloud
interactions in OSCAR (Sect. 2.9.2). Here we show the simulated radiative
forcing as a function of the total burden of soluble aerosols (left-hand
side) or of the change in that burden since preindustrial times (right-hand side).
In the former case, the grey dotted lines show the preindustrial burden we
calculate; in the latter, the red area shows the 90 % range of RF provided by
The radiative forcing induced by BC deposition on snow is taken directly
proportional to the regional BC emissions. It is parameterized by a global
radiative efficiency with respect to emissions
(
The global radiative efficiency with respect to emissions
The radiative forcing induced by changes in land cover is modeled following
the first-order equation of
The upward transmittance is set to
The climate module of OSCAR is relatively simple compared to other
models, as the energy budget is done only on a global scale and no water budget
is explicitly done. The two-box model used to estimate global temperature
change is parameterized by – among other things – an equilibrium climate
sensitivity. Therefore, all the feedbacks occurring within the climate
system, such as changes in tropospheric water vapor, in ice cover, or in cloud
cover are implicitly accounted for in a linear manner and with the
assumption that all climate forcers induce the same level of feedback.
Similarly, in this version of OSCAR, the regional patterns of temperature or
precipitation changes are the same whatever the climate forcer. Also, no
non-radiative biophysical or physiological effect relative to the land
vegetation – e.g., a change in evapotranspiration induced by a change in
land cover or in atmospheric CO
The first step to calculate global warming is to calculate global radiative
forcing. So as to ease the notations, following
To estimate global warming, however, we have to account for the so-called
“efficacy” of these forcings, i.e., we have to introduce new parameters
(
Now, to estimate global precipitation change, we also need to estimate how
much of this top-of-the-atmosphere RF is actually occurring within the
atmosphere – thus creating a local energy imbalance – in opposition to the
RF occurring at the Earth's surface. To do so, we introduce new parameters
that quantify this atmospheric fraction for several groups of forcers: carbon
dioxide alone (
Similarly to what is done in other simple models (e.g.,
The two-box model used to model the global surface temperature change has two
state variables: the global surface temperature itself (
The first set of parameters of this module, for global temperature, can be
calibrated on 25 CMIP5 global circulation models. First, using
outputs from the “abrupt4xCO2” and “piControl” experiments, we estimate the
steady-state temperature change at quadrupled CO
The second set of parameters, for the pattern scaling, is calibrated on the
same CMIP5 model chosen for the global temperature response. This pattern
scaling can be based on the quadrupled CO
Changes in global yearly precipitation (
The first set of parameters of this module, for global precipitation, can be
calibrated on 25 CMIP5 global circulation models, chosen
independently from the one used for the calibration of the temperature
module. Using outputs from the “abrupt4xCO2” and “piControl” experiments, we
calibrate the two parameters of Eq. (82) thanks to a linear fit with a
constant term made between the global surface temperature and global
precipitation. The constant term is assumed to correspond to the RF term,
since the radiative forcing is actually constant in the quadrupled CO
The second set of parameters, for the pattern scaling, are also calibrated on
the same CMIP5 model as the global precipitation response. The
The ocean heat content (OHC) – a third climate change indicator – is simply
deduced from the two-box model used for the temperature. However, we need to
introduce a coefficient (
When put together, all previous equations from Eqs. (1) to (84) form a system of
ordinary differential equations of first order, for a subset of the variables
of the model. These variables are the state variables – or prognostic
variables – of the dynamical system described by the differential equations.
They are compiled in Table 1, along with the drivers of the model. By
definition, knowledge of both the drivers and the state variables, at any
time step, gives knowledge of all the other variables of the system, at that
time step. These other secondary variables – or diagnostic variables – are
compiled in Table 2. The differential system is solved with the forward Euler
method
List of drivers and state variables of the model.
List of secondary variables of the model.
We make two series of historical simulations, with the goal of evaluating the
performance of each module of OSCAR v2.2 separately and of the fully coupled
model itself. The simulations are realized within a probabilistic framework:
a set a drivers and parameters is drawn randomly, with equiprobability, from
the pool of potential driving datasets and parameterizations that is
summarized in Table 3. With the given drivers and parameters, two simulations
are made: one in which the atmospheric concentrations of well-mixed
greenhouse gases, the total and per component radiative forcings, and the
various climate variables are prescribed to the model; and another in which
nothing more than the drivers is prescribed. The first simulation is called
“offline”, and the second “online”. The offline simulation has the interest
of uncoupling the different modules of OSCAR, thus separating them from each
other and allowing an easier diagnosis of any potential issue or bias in each
module. The online simulation is meant to diagnose the behavior of OSCAR when
it is used as a proper Earth system model, i.e., when it is driven only by the
anthropogenic perturbations of the system. The Monte Carlo ensemble size is
10 000 simulations which are drawn from a pool of more than 10
List of driving datasets and parameterizations for the probabilistic setup of the model. The “#” column shows how many options are available for the given parameter or set of parameters. Superscripts are omitted for clarity.
Continued.
As described in Sect. 2.3.2, the disaggregation of the terrestrial
biosphere follows the nine regions of
The following sections are dedicated to discussing the results of the historical simulations for the main variables of the model. Each section refers to one of Figs. 5 to 12. In the case of the offline simulation, we show and discuss the “reconstructed” time series of those variables that are prescribed to the model. In other words, in the following, the offline atmospheric growth rate and concentration of a given WMGHG are reconstructed as the balance of the prescribed emissions and the simulated fluxes. The offline RFs are reconstructed on the basis of the reconstructed atmospheric concentrations. The climate variables, however, are reconstructed on the basis of the prescribed RFs, so that we can discuss the performance of the climate module alone, i.e., when it is not coupled to any other module.
The median land-use change emissions simulated by the book-keeping module of
OSCAR are of the same order of magnitude – though smaller than – the values
reported by the global carbon project
Results of our simulations with OSCAR, for carbon
dioxide. The offline simulation is shown in blue, and the online simulation
in black. Other colors are references we compare our results to. The
left-hand panels show the time series from 1900 to 2010, the thick colored
lines indicate the median of the ensemble of simulations, and the colored
area its 5th to 95th percentiles. The right-hand panels show the probability
distribution function (PDF) from the ensemble of simulations, for the
averaged last 10 years of simulation. Reference for the first three fluxes is
the GCP
The median land sink we simulate in the offline simulation is slightly
smaller (in absolute value) than the estimate by
The median ocean sink OSCAR simulates matches relatively well the estimate by
In both the online and offline simulations, the simulated atmospheric growth
rate is very close, on average, to the one reported by NOAA/ESRL
Finally, regarding excess atmospheric CO
The emissions from biomass burning are shown and discussed here, despite
being mainly a product of the carbon cycle in OSCAR, since they are part of
the atmospheric balance of methane. One can see that our approach of
calculating these emissions endogenously gives values of the same order of
magnitude as those of
Results of our simulations with OSCAR, for methane, with the same
format as for carbon dioxide. References are ACCMIP
When compared to the multi-model mean of WETCHIMP
The median lifetime of methane with regard to the OH sink which we simulate
is very close to the best-guess value of
In the online simulation the median atmospheric growth rate of methane we
simulate is close to the observed one, over the short period of observation
we have at our disposal. OSCAR manages to reproduce the slowdown of
atmospheric increase around the year 2000; this slowdown is mainly driven by
anthropogenic emissions in our model. After 2005, however, the atmospheric
growth resumption is too fast when compared to observations. In the offline
simulation the picture is completely different: the atmospheric growth rate
– reconstructed as the balance between the concentration-driven sinks and
the anthropogenic emissions normally driving OSCAR in online mode – is
systematically higher than in the online case, by 10 to 20 MtC yr
Regarding atmospheric CH
The nitrous oxide emissions from biomass burning are shown here mainly to
point out that they are strictly similar to that of methane in Fig. 6. This
is true for all non-CO
Results of our simulations with OSCAR, for nitrous oxide, with the
same format as for carbon dioxide. References are
The median lifetime of nitrous oxide with regard to the stratospheric sink
which we simulate is very close to the best-guess value of
On average, the median atmospheric growth rate we simulate is close to the
observed one over 1979–2010, although slightly smaller for the offline
simulation. The observed variability, however, is not reproduced by our
model, be it in the online or offline setup. This suggests that a biological
process related to nitrous oxide is missing in our model. Processes such as
biological production in terrestrial or aquatic systems are viable candidates
In the online simulation, the excess atmospheric concentration we simulate is
lower than the one observed: the median is actually parallel to the
observations with a distance of
While other species are shown in Fig. S50, here we show only the first
compound of each group of halogenated compounds (i.e., HFC-23 for HFCs, CF
Results of our simulations with OSCAR, for halogenated
compounds, with the same format as for carbon dioxide. Reference for the atmospheric
concentrations is
Results of our simulations with OSCAR, for ozone, with the same
format as for carbon dioxide. Reference for the global burden is
If we look at the variables that summarize the two effects of the halogenated
compounds within the climate system, i.e., effective equivalent stratospheric
chlorine and radiative forcing, we can have an overview of the performance of
this module. Regarding the EESC simulated by our model, it is lower than the
one calculated on the basis of the
Regarding tropospheric ozone, the median change in burden simulated by OSCAR
is very close to the only point in time we have from the
Regarding stratospheric ozone, our slightly underestimated EESC induces a
slightly underestimated change in column burden (in absolute value), again
over the reference period 1850–2000. Nonetheless, the estimate by the
Regarding the direct effect of aerosols, OSCAR's ability to match the IPCC
best guess
Results of our simulations with OSCAR, for
aerosols, with the same format as for carbon dioxide. Reference 1 for the
radiative forcing and its 90 % uncertainty range is IPCC
The cases of POA and BC are very comparable: our median RFs are significantly smaller (in absolute value) than the IPCC references, and the distributions are close to a log-normal one and with a relatively consistent spread. For both aerosols, however, if we remove the contribution of biomass burning aerosols to the IPCC best guesses, our median estimates are much closer. This odd feature does not greatly affect the overall performance of the model (see next section), as the IPCC best-guess estimate for combined biomass burning POA and BC is zero. It strongly suggests, however, that the way these biomass burning aerosols are treated in OSCAR can be improved.
In the case of nitrate, our median RF is relatively close to the IPCC best guess, whereas our distribution does not go as far in the negative values as the IPCC uncertainty range, as a result of having too few – only two – possible parameterizations for these aerosols. In the case of SOA, our median RF is very small, owing to the fact that one out of three simulations has the SOA turned off, and the distribution clearly shows that we only have three possible parameterizations for this aerosol. Also, because all the radiative efficiencies of SOA available to OSCAR are negative, the only way it could go into the positive-value domain would be to have varying biogenic emissions of NMVOCs, which is not the case in this version.
Regarding the cloud effect of aerosols, which includes both the so-called
semi-direct and indirect effects, OSCAR performs well and its median estimate
meets the IPCC best guess in 2010. This is mostly due to the way this effect
is calculated in our model, as the main sensitivity parameter of the module
(i.e.,
When we combine together the RF induced by all well-mixed greenhouse gases,
we see that the median of both our online and offline simulations are
slightly higher in 2010 than the estimate by
Results of our simulations with OSCAR, for radiative forcing, with
the same format as for carbon dioxide. Reference for the radiative forcing
and its 90 % uncertainty range is IPCC
Regarding the two RFs induced by surface albedo, our two simple modules
simulate values that meet the IPCC estimate for the year 2010. For black
carbon deposition on snow, this could be expected from our rescaling of the
global sensitivity parameter
All in all, the total RF simulated by OSCAR – which is the sum of the above
four RFs and the three drivers prescribed directly as radiative forcing –
has a median value in the year 2010 close to the IPCC best guess, but
slightly higher. In the online case it has a relatively consistent spread,
whereas in the offline one the spread is much larger. This large spread is
dominated by the large spread in the RF of WMGHGs which itself is dominated
by the large spread in offline atmospheric CO
Global mean surface temperature, which is our prime proxy of climate change, is relatively well simulated by OSCAR over the 1900–2010 period. We note, however, that the 1940s warmer period is not reproduced, and during the last 10 years of simulation the simulated temperature tends to be higher than the observations. Interestingly, OSCAR simulates a slowdown of the warming during these last 10 years – the so-called hiatus period. The fact that the slowdown is simulated in both the offline and online setups suggests it is a feature of our climate module alone, and thus can be explained by the RFs we use as inputs. However, the lack of inter-annual variability in OSCAR makes any further investigation on the topic virtually impossible. Note also that the offline simulation gives a narrower range than the online one because only one set of radiative forcings is prescribed in the former case.
Results of our simulations with OSCAR, for climate, with the same
format as for carbon dioxide. Reference 1 is HadCRUT4
As for the global sea surface, one can see here the limits of our pattern-scaling approach: the single proportionality parameter makes the time series
of sea surface temperature homothetic to that of global surface temperature.
If the simulated temperature follows relatively well the observations over
1900–2010, the simulated temporal variability does not match the observed
one. Similarly, the simulated local surface temperatures, shown in Fig. S51, are proportional to the global one, which gives temperature changes
consistent with the CRU dataset
Although we cannot compare our global yearly precipitation with a long enough time series of observation, we can note that OSCAR simulates a wide range of precipitation changes, with a non-negligible difference between the offline and online configurations. This is mostly caused by the difference between the simulated RF of aerosols in the online setup and the prescribed RF in the offline one. Regarding local yearly precipitation, shown in Fig. S52, OSCAR does not manage to capture the past variation of this variable, in any of our regions. This has limited impact on the model's results, since in Sect. 2.3.2 we calibrate the sensitivity parameters of NPP and heterotrophic respiration in two steps, the first of which being driven by temperature alone. It does, however, impact our simulated methane emissions from wetlands (see above). More work is needed to improve that aspect of the model.
Finally, the ocean heat content simulated with our model is of the right
order of magnitude, logically owing to the good simulated RF and temperature.
It follows relatively well the variations of the observations for both online
and offline simulations, except over the last 10 years of simulation. This
could be explained by our choice of a single value for
In this paper, we have provided a complete description of the compact Earth
system model OSCAR v2.2, and we have presented the model's results in the
case of an historical simulation. Overall, despite some caveats discussed in
the previous section, we conclude that the model performance is good,
especially given its level of complexity. OSCAR manages to satisfactorily
reproduce most of the past changes in the global Earth system, with an even
better performance over the recent period for which better driving data are
available. However, we note that a good performance of a simple model over
the historical period does not warrant a good performance in any other
simulation. In the case of OSCAR, since its parameters are generally
calibrated on simulations that go relatively far from the historical
conditions (e.g., under quadrupled CO
The fact that OSCAR has been developed to be used in a probabilistic setup is
an additional strength of the model, although the spread in the model's
results for some components may greatly differ from the uncertainty range
assessed by studies based on more complex models and/or observations. In
addition to the reasons discussed in the section above, there are two more
general causes to that feature, owing to the principles underpinning OSCAR's
development (expounded in Sect. 2.1). First, because all the modules of
OSCAR interact with each other, the model's overall causal chain is fairly
complex (as illustrated in Fig. 1) and it has many degrees of freedom –
actually more than most CMIP5 complex Earth system models. These many degrees
of freedom increase the odds of seeing a given simulation depart unreasonably
from the plausible range of results. Second, OSCAR is not designed to emulate
a given complex Earth system model as a whole: each of its modules is
essentially an emulator, and OSCAR is the combination of these emulators.
Consequently, in a given parameterization, two modules could emulate the
sensitivities of two complex models that are physically inconsistent with one
another (e.g., the implicit ocean transport of the climate module could be
inconsistent with that of the carbon-cycle module), therefore potentially
leading to unreasonable results. These two elements explain why OSCAR's
average or median simulation can differ from the average of a model
intercomparison exercise we used for calibration, and why the model's results
can show very large spreads. A way to solve this and improve the
probabilistic setup is to use observational constraints, either to rate a
given parameterization and therefore give it a lower weight if its too far
from the observations
To conclude, we want to suggest a few tracks for future development of the
model. Despite its overall good performance, the model can indeed be
improved, especially in terms of consistency of modeling. We see three broad
aspects of the model for which such improvements would be advisable. First, the carbon cycle can be improved by
inclusion of nutrient limitations for the land carbon cycle, and of the
biological pump for the ocean carbon cycle. Inclusion of the nitrogen cycle
would couple the carbon cycle and the atmospheric chemistry, as the carbon
sinks would be affected by deposition of active nitrogen that would be
induced by NO
The source code of this version of OSCAR is available upon request to the corresponding author. Detailed information as to the complex model data processing can also be provided upon request. A brief user manual is provided with the code.
All the data presented in this paper can be obtained upon request to the corresponding author.
Version 2.1 of OSCAR is completely described by
The main changes between v2.1 and v2.2 are the following: development of the ocean carbon-cycle module to include the stratification effect calibrated on CMIP5 models, extension of the terrestrial carbon-cycle module to be calibrated on many TRENDY and CMIP5 models, creation of a wildfire module, extension of the wetland module to be calibrated on many WETCHIMP models, development of the stratospheric sink module to include the effect of ozone-depleting substances and age-of-air change, development of the tropospheric ozone module to include a regionalization and the effect of climate change, development of the stratospheric ozone module to include the effect of nitrous oxide and climate change, development of the aerosols module to have explicit and regionalized parameterizations, creation of the surface albedo modules, and development of the climate module to include a global precipitation response.
Many other small and specific changes were also made during the development of the latest version.
Version 2.0 of OSCAR is exactly the same as version 2.1, with two significant
exceptions. First, non-CO
It can also be noted that the main change between the previous versions of OSCAR and v2.0 is the computing language used to code the model. While previous versions were coded in Scilab, the following versions (i.e., from v2.0 onward) have been coded in Python.
Version 1.1 of OSCAR is an update of version 1.0, described by
Version 1.0 of OSCAR is described by
These models are those whose outputs we use to calibrate some of OSCAR's parameters. In other words, we do not list here the models for which we simply read OSCAR's parameter value in, e.g., a table of another study. Note that here we give the models' name as given by the study we base our calibration on. These names may vary across studies and from the official name itself.
For the ocean carbon cycle, stratification effect (Sect. 2.3.1): CESM1-BGC, IPSL-CM5A-LR and MPI-ESM-LR.
For the land carbon cycle, transient response of net primary productivity and heterotrophic respiration (Sect. 2.3.2): BCC-CSM1.1, CESM1-BGC, CanESM2, HadGEM2-ES, IPSL-CM5A-LR, MPI-ESM-LR and NorESM1-ME.
For the land carbon cycle, transient response of wildfires (Sect. 2.3.2): CESM1-BGC, IPSL-CM5A-LR, MPI-ESM-LR and NorESM1-ME.
For the atmospheric burden of sulfate, primary organic and black carbon aerosols (Sect. 2.9.1): CSIRO-Mk3.6.0, GDFL-CM3 and MIROC-CHEM.
For the atmospheric burden of secondary organic aerosols (Sect. 2.9.1): GFDL-CM3.
For the indirect effect of aerosols (Sect. 2.9.2): CSIRO-Mk3.6.0, and IPSL-CM5A-LR.
For the climate module, both the temperatures and the precipitation (Sect. 2.11.2 and 2.11.3): ACCESS1.0, ACCESS1.3, BCC-CSM1.1, BCC-CSM1.1m, CanESM2, CCSM4, CNRM-CM5, CNRM-CM5.2, CSIRO-Mk3.6.0, GFDL-CM3, GFDL-ESM2G, GFDL-ESM2M, GISS-E2-H, GISS-E2-R, HadGEM2-ES, IPSL-CM5A-LR, IPSL-CM5A-MR, IPSL-CM5B-LR, MIROC5, MIROC-ESM, MPI-ESM-LR, MPI-ESM-MR, MPI-ESM-P, MRI-CGCM3 and NorESM1-M.
For the terrestrial carbon cycle, preindustrial net primary productivity and heterotrophic respiration (Sect. 2.3.2): CLM4.5, JSBACH, JULES, LPJ, LPJ-GUESS, LPX-Bern, OCN, ORCHIDEE and VISIT.
For the terrestrial carbon cycle, preindustrial wildfires (Sect. 2.3.2): CLM4.5, JSBACH, LPJ, LPJ-GUESS, ORCHIDEE and VISIT.
For the natural wetlands, preindustrial state (Sect. 2.5.2): CLM4-Me, DLEM, IAP-RAS, LPJ-Bern, LPJ-WSL, ORCHIDEE and SDGVM.
For the natural wetlands, transient response of the area extent (Sect. 2.5.2): CLM4-Me, DLEM, LPJ-Bern, ORCHIDEE, SDGVM and UVic-ESCM.
For the stratospheric sink, transient response of the age of air (Sect. 2.6.1): AMTRAC3, CAM3.5, CMAM, Niwa-SOCOL, SOCOL, ULAQ and UMUKCA-UCAM.
For the stratospheric ozone, transient response to chlorine and climate change (Sect. 2.8.2): AMTRAC3, CCSR-NIES, CMAM, CNRM-ACM, LMDZrepro, MRI, Niwa-SOCOL, SOCOL, ULAQ, UMSLIMCAT and UMUKCA-UCAM.
For the tropospheric ozone, transient response to precursors emissions (Sect. 2.8.1): CICERO-OsloCTM2, NCAR-CAM3.5, STOC-HadAM3 and UM-CAM.
For the tropospheric ozone, transient response to climate change (Sect. 2.8.1): CESM-CAM-superfast, GFDL-AM3, GISS-E2-R, MIROC-CHEM, MOCAGE, NCAR-CAM3.5, STOC-HadAM3 and UM-CAM.
For the atmospheric burden of sulfate, primary organic and black carbon aerosols (Sect. 2.9.1): GISS-E2-R.
For the atmospheric burden of secondary organic aerosols (Sect. 2.9.1): GISS-E2-R.
For the indirect effect of aerosols (Sect. 2.9.2): GFDL-AM3, GISS-E2-R, HadGEM2, MIROC-CHEM and NCAR-CAM5.1.
First of all, we thank all the modeling team that produced the data we processed to build OSCAR. We also thank those who shared data or info from their model or study: V. K. Arora, P. Cadule, A. R. Douglass, T. Fu, W. Fu, C. D. Jones, J.-F. Lamarque, Y. Lee, S. Levis, K. Lindsay, U. Lohmann, J. R. Melton, V. Naik, J. Pongratz, J. T. Randerson, B. Ringeval, L. Rotstayn, X. Shi, D. T. Shindell, S. Sitch, B. D. Stocker, J. Tjiputra, N. Viovy, S. Watanabe, and H. Yu. Development of versions 2.0 and 2.1 was part of the ACACCYA project funded by the GIS Climat-Environnement-Société. Development of version 2.2 was supported by the European Research Council Synergy project IMBALANCE-P (grant ERC-2013-SyG-610028). Thomas Gasser also acknowledges support from the Research Council of Norway with a visiting researcher grant (#249972) during the writing of this paper. Download and processing of CMIP5 data was made on the IPSL Prodiguer-Ciclad facility which is supported by CNRS, UPMC, and Labex L-IPSL, and funded by the ANR (grant #ANR-10-LABX-0018) and the European FP7 IS-ENES2 project (grant #312979). Download of ACCMIP and CCMVal2 data was made on the BADC facility which is part of the NERC-NCAS. Edited by: G. A. Folberth Reviewed by: two anonymous referees