We document the first version of the Centre National de Recherches Météorologiques Earth system model (CNRM-ESM1). This model is based
on the physical core of the CNRM climate model version 5 (CNRM-CM5) model and employs the Interactions between Soil, Biosphere and Atmosphere (ISBA) and the
Pelagic Interaction Scheme for Carbon and Ecosystem Studies (PISCES) as terrestrial and oceanic
components of the global carbon cycle. We describe a preindustrial and
20th century climate simulation following the CMIP5 protocol. We detail
how the various carbon reservoirs were initialized and analyze the behavior
of the carbon cycle and its prominent physical drivers. Over the 1986–2005
period, CNRM-ESM1 reproduces satisfactorily several aspects of the modern
carbon cycle. On land, the model captures the carbon cycling through
vegetation and soil, resulting in a net terrestrial carbon sink of 2.2 Pg C year
Earth system models (ESMs) are now recognized as the current state-of-the-art models (IPCC, 2013), expanding the numerical representation of the climate system of the 4th Assessment Report (IPCC, 2007). They enable the representation of subtle nonlinear interactions and feedbacks of different magnitude and signs of various biogeochemical and biophysical processes with the climate system. The latter contribute, in addition to the atmospheric radiative properties and global climate dynamics, to determining the Earth's climate variability (Arora et al., 2013; Cox et al., 2000; Friedlingstein and Prentice, 2010; Schwinger et al., 2014; Wetzel et al., 2006).
Although there is no uniformly accepted definition, ESMs generally bring together a global physical climate model and land and ocean biogeochemical modules (Bretherton, 1985; Flato, 2011). As such, they enable the representation of the global carbon cycle. The models of this class have played a larger role in the 5th IPCC report than in previous reports, primarily through their contribution to the concentration- and emission-driven experiments that compose CMIP5.
Even if the concept of Earth system modeling is being extended to include
further processes and reservoirs (e.g., nitrogen cycle, aerosols) (Hajima et
al., 2014), there are still large uncertainties in the representation of the
carbon cycle and its interactions with climate (Anav et al., 2013a;
Friedlingstein et al., 2013; Piao et al., 2013). To reduce them, there is a
need for improvements of both physical and ecophysiological parameterizations
(Dalmonech et al., 2014), and for the development of observation-based
methods to constrain model projections (Wenzel et al., 2014). But the
reduction of carbon cycle–climate uncertainties also requires a greater
number and diversity of ESMs. This path is promoted and followed by various
international initiatives like the Global Carbon Budget
(
This article documents the first IPCC-class ESM developed at Centre National de Recherches Météorologiques (CNRM) and provides a basic evaluation of the model's skill. This model is based on the CNRM-CM5.1 climate model jointly developed by CNRM and Cerfacs (Centre Européen de Recherche et de Formation Avancée en Calcul Scientifique), which has contributed to the 5th phase of the Coupled Model Intercomparison Project (CMIP5) (Voldoire et al., 2013). CNRM-CM5.1 did not include a representation of the global carbon cycle but accounted for chemical–climate interactions with an interactive stratospheric chemistry module (Cariolle and Teyssèdre, 2007). While this configuration of CNRM-CM5 contributed to the CMIP5 results publicly released, a first intermediate version of the CNRM ESM was developed with the inclusion of the marine biogeochemistry model Pelagic Interaction Scheme for Carbon and Ecosystem Studies (PISCES) (Aumont and Bopp, 2006). This model version was evaluated against modern oceanic observations (Séférian et al., 2013) and employed in various studies (Frölicher et al., 2014; Laufkötter et al., 2015; Schwinger et al., 2014; Séférian et al., 2014).
A terrestrial carbon cycle module has been under development at CNRM since the 2000s (Calvet and Soussana, 2001; Calvet et al., 2008, 2004; Gibelin et al., 2008, 2006), but it has never been coupled to an atmosphere–ocean model. This carbon cycle module evolved from the physically based Interactions between Soil, Biosphere and Atmosphere (ISBA) model (Noilhan and Mahfouf, 1996; Noilhan and Planton, 1989) and is able to simulate the surface carbon fluxes and the terrestrial carbon pools. The carbon fluxes module was extensively tested over France and Europe (Sarrat et al., 2007; Szczypta et al., 2012), and the carbon cycle module was tested for temperate and high-latitude regions (Gibelin et al., 2006, 2008) and was used more recently in studies of carbon cycling over the Amazon basin (Joetzjer et al., 2015, 2014), permafrost regions (Rawlins et al., 2015) and at global scale (Carrer et al., 2013b). In this work, this terrestrial carbon cycle module is coupled to a global climate model for the first time.
Here, we present a first evaluation of the CNRM-ESM1. In Sect. 2, we describe the model, focusing on the Earth system's components and aspects of the climate model that are particularly relevant to the global carbon cycle. We describe in Sect. 3 the preindustrial control and 20th century experiments that we conducted, together with the forcings used and how the experiments were initialized. In Sect. 4, we present and discuss the results of these experiments. We summarize the results in Sect. 5 and present conclusions.
CNRM-ESM1 is based on the physical core of the CNRM-CM5.1 atmosphere–ocean general circulation model extensively described in Voldoire et al. (2013), which accounts for the physical and dynamical interactions occurring between atmosphere, land, ocean and sea ice.
The atmospheric component is based on version 6.1 of the global spectral
model ARPEGE-Climat, which corresponds to an updated version of the
atmospheric code used in CNRM-CM5.1. This updated version of the atmospheric
code is derived from cycle 37 of the ARPEGE-IFS (integrated forecast system)
numerical weather prediction model developed jointly by Météo-France
and the European Center for Medium-range Weather Forecast. In CNRM-ESM1, the
geometry, parameterizations and dynamics have been chosen to match the
choices made for CNRM-CM5.1. Thus, differences are mainly due to debugging
and recoding. The atmospheric physics and dynamics are solved on a T127
triangular truncation that offers a spatial resolution of about
1.4
The land-surface component is an updated version of the SURFface EXternalisée modeling platform (SURFEXv7.3) (Masson et al., 2013b) associated with the total runoff integrating pathways (TRIP) river routing model (Oki and Sud, 1997). SURFEX was designed so that the same code could be run offline or coupled to a general circulation model (GCM), allowing for easy transfer from offline improvements to the coupled model and to be able to compare online and offline runs.
This model prognostically computes the exchange of energy, water and carbon between the atmosphere and three types of natural surfaces: land, free water bodies and oceans or seas. The energy, water and carbon balances are calculated separately for each surface type and area averaged over each atmospheric grid cell. The natural land surfaces are represented by the module originally developed by Noilhan and Planton (1989). This module solves the surface energy and soil water budgets using the force–restore method and a composite soil–vegetation–snow approach. The version used here is the same as for CNRM-CM5.1; e.g., the soil hydrology uses three vertical layers (Boone et al., 1999) while soil temperature is solved using four vertical layers. In CNRM-ESM1, land-surface albedo benefits from an improved spatial representation derived from MODIS satellite measurements (Carrer et al., 2013a) except for the area covered by snow for which the albedo is prognostically computed following Douville et al. (1995). Over water bodies and oceans, we use the CNRM-CM5.1 parameterization for momentum and energy fluxes except for the sea-to-air turbulent fluxes that are computed from the Coupled Ocean–Atmosphere Response Experiment (COARE) scheme (Fairall et al., 2003). Interactions between the land-surface energy and water budgets and the terrestrial carbon cycle module are detailed in Sect. 2.3.1.
The ocean component uses version 3.2 of the Nucleus for European Modelling of
the Ocean (NEMO) model (Madec,
2008) in the ORCA1L42 configuration. This configuration offers a horizontal
resolution from 1 to
The sea-ice model used in CNRM-ESM1 is Global Experimental Leads and ice for
ATmosphere and Ocean (GELATO6).
This model employs the same horizontal grid as NEMO and solves sea-ice
dynamics and thermodynamics every 6 h. This model represents an updated
version of the former sea-ice model used in CNRM-CM5.1 (Voldoire et al.,
2013). In GELATO6, sea-ice dynamics is computed using the elastic
viscous–plastic scheme proposed by Hunke and Dukowicz (1997) formulated on
an Arakawa C-grid (Bouillon et al., 2009). To simulate the response of sea
ice to convergence–divergence movements, GELATO6 employs a redistribution
scheme derived from Thorndike et al. (1975). This scheme ensures the
representation of the rafting phenomenon for the slab of sea ice thinner than
0.25 m and of ridging for the slab thicker than 0.25 m. GELATO6 includes a
thermodynamic scheme that resolves the evolution of four ice thickness
categories (0–0.3, 0.3–0.8, 0.8–3 and over 3 m). These four slabs of sea
ice are modeled with 10 vertical layers unevenly distributed across the slab
thickness. An enhanced resolution at the top of the slab is used to better
represent the evolution of sea ice in response to the high-frequency
variability of the atmospheric thermal forcing. Besides, all sea-ice slabs
may be covered with one snow layer. In GELATO6, the snow layer is considered
to occult the transfer of light across the snow–sea ice–ocean continuum.
This snow layer can age or form ice using the formulation described in Salas
y Mélia (2002). Since CNRM-CM5.1, the coupling between NEMO and GELATO
has been revised in order to improve the conservation of water and salt. In
the previous model version, CNRM-CM5.1, there was a large drift in salinity
(
In CNRM-ESM1, exchanges of momentum, water and energy between the atmosphere and the surface models occurs every atmospheric time step (i.e., 30 min) because SURFEX is a submodel of the atmospheric code. The coupling between the atmosphere and the ocean models is handled by the OASIS coupler (Valcke, 2013) and occurs every 6 h. In CNRM-ESM1, the frequency of coupling between the ocean and atmosphere models has been increased compared to CNRM-CM5 in order to better resolve the dynamics of the sea ice, which is resolved at this time step (i.e., 6 h).
The atmospheric chemistry scheme in CNRM-ESM1 consists of an interactive
linear ozone chemistry performed with the MOdèle BIDImensionnel de Chimie
(MOBIDIC, Cariolle and Teyssèdre, 2007) including a representation of the three-dimensional
atmospheric CO
As in CNRM-CM5, the ozone mixing ratio is treated as a prognostic variable
with photochemical production and loss rates climatology computed by a full
chemistry scheme. That is, the net photochemical production in the ozone
continuity equation is solved using a first-order Taylor series around the
local value of the ozone mixing ratio, air temperature and the overhead ozone
column. Ozone destruction terms are used to parameterize the heterogeneous
chemistry as a function of the equivalent chlorine content prescribed for the
actual year. All Taylor coefficients of this linearized scheme were
determined using a two-dimensional chemistry scheme with 56 constituents, 175
chemical reactions, and 51 photoreactions (Cariolle and Brard, 1985).
Photochemical production and loss rates of ozone rely on the main gas-phase
reactions driving the NO
In CNRM-ESM1, the atmospheric CO
In CNRM-ESM1, the interactions between climate and vegetation are handled by the ISBA scheme embedded in the SURFEX (Surface Externalisée) model. The land biogeochemical module in ISBA represents land-surface physics, plant physiology, carbon allocation and turnover, and carbon cycling through litter and soil (Calvet and Soussana, 2001; Calvet et al., 1998; Gibelin et al., 2006, 2008). The land cover is represented by nine plant functional types (PFT; given in Fig. 1) and three non-vegetated surface types that are determined spatially by the ECOCLIMAP physiographic database (Masson et al., 2013a).
Fraction of dominant vegetation type as prescribed in SURFEX. This
fraction results from aggregation of the various ECOCLIMAP's vegetation
types at 1 km resolution over the T127 CNRM-ESM1 horizontal grid
(
ISBA uses a semi-mechanistic treatment of canopy photosynthesis and mesophyll conductance following the Jacobs et al. (1996) and Goudriaan et al. (1985) photosynthesis model. Mesophyll conductance in this framework corresponds to the rate of photosynthesis under light-saturated conditions (Jacobs et al., 1996). As such, this scheme does not explicitly account for Michaelis–Menten kinetics of the Rubisco enzyme found in Farquhar et al. (1980) and Collatz et al. (1992) models. ISBA includes a representation of the soil water stress. Key parameters of the photosynthesis model respond to the soil water stress, permitting the representation of drought-avoiding and drought-tolerant responses to drought. For low vegetation and for trees, the response to drought is based on the meta-analyses of Calvet (2000) and Calvet et al. (2004), respectively.
The model simulates a ratio of intercellular CO
ISBA simulates the evolution of six reservoirs of biomass including leaf,
wood and roots, and assumes the existence of metabolic/structural
reservoirs of biomass (Gibelin et al., 2008). Vegetation biomass is
simulated interactively based on the carbon assimilated by photosynthesis,
and decreased by turnover and respiration. The autotrophic respiration
combines the respiration from all these reservoirs except the woody
reservoir that is supposed not to respire (Gibelin et al., 2008). In
this model, the vegetation phenology results directly from the carbon
balance of the leaves. Therefore, phenology is completely driven by
photosynthesis and no growing degree-day model is used. A key advantage of
this approach is that most of the soil and atmospheric drivers (the abiotic
drivers) of phenology are accounted for without any additional parameters
(Szczypta et al., 2014). Leaf area index (LAI) is determined from the
leaf biomass and the specific leaf area index, which varies as a function of
leaf nitrogen concentration and plant functional type (Gibelin et al.,
2006). ISBA uses an implicit nitrogen limitation parameterization, which is
based on the meta-analysis of leaf nitrogen measurement under CO
The soil organic matter and litter module in ISBA follows the soil carbon
part of the CENTURY model (Parton et al., 1988). Four pools of litter
are represented. They are differentiated by their location above- or
belowground and their content of lignin. The litter pools are supplied by
the fluxes of dead biomass from each biomass reservoir (turnover) as
described in Gibelin et al. (2008). The three soil organic matter
reservoirs (active, slow and passive) are characterized by their resistance
to decomposition with turnover times spanning from a few months for the
active pool to 240 years for the passive pool. Heterotrophic respiration and
hence the flux of CO
Changes in the carbon balance of the vegetation affect the energy and water balance, and hence the climate, through changes in stomatal conductance and LAI. Through its control on leaf transpiration, stomatal conductance affects latent heat flux and the surface energy balance. LAI on the other hand affects evapotranspiration because it is used to scale leaf-level to canopy-level transpiration and evaporation from the interception reservoir (water intercepted by leaves).
In CNRM-ESM1, except for crops, changes in LAI do not affect the albedo of the land surface, as it is the case in some other models. As mentioned earlier, albedo is derived from satellite observations corrected in the presence of snow, but does not depend on the changes in LAI calculated by the model. This limits the biophysical feedback from vegetation change to the atmosphere.
The ocean biogeochemical model of CNRM-ESM1 is PISCES (Aumont and Bopp, 2006). This model simulates the biogeochemical cycles of oxygen, carbon and the main nutrients with 24 state variables. Macronutrients (i.e., nitrate and ammonium, phosphate, silicate) and micronutrients (i.e., iron) ensure a better representation of the phytoplankton dynamics, because these five nutrients contribute to the nutrient limitation process (Aumont et al., 2003). PISCES represents two size classes of phytoplankton (i.e., nanophytoplankton and diatoms). Dependence of growth on temperature is parameterized according to Eppley et al. (1969). Growth rate is also limited by the external availability in nutrients using the Michaelis–Menten relationships. Diatoms differ from nanophytoplankton by their need in silicon, by higher requirements in iron (Sunda and Huntsman, 1997) and by higher half-saturation constants because of their larger mean surface-to-volume aspect ratio. Zooplankton is represented by two size classes: microzooplankton and mesozooplankton.
PISCES can be considered as a Monod model (Monod, 1942) since it does not represent the internal concentration of nutrients in the cells. The ratios between carbon, nitrate and phosphate are kept constant to the values proposed by Takahashi et al. (1985) in all living and non-living pools of organic matter. However, internal concentrations of iron in both phytoplankton and of silicon in diatoms are prognostically simulated. They depend on the external concentration of these nutrients, on the potential limitation by the other nutrients and on light availability.
Phytoplankton chlorophyll concentration is prognostically simulated
following Geider et al. (1998). PISCES simulates semi-labile dissolved
organic matter, small and big sinking particles, which differ by their
sinking speeds (i.e., 3 m d
The boundary conditions account for nutrient supply from three different sources: atmospheric dust deposition for iron and silicon (Jickells and Spokes, 2001; Moore et al., 2004; Tegen and Fung, 1995), rivers for carbon (Ludwig et al., 1996) and sediment mobilization for sedimentary iron (de Baar and de Jong, 2001; Johnson et al., 1999). In CNRM-ESM1, riverine input of carbon has been revised from Ludwig et al. (1996) in accounting for the interannual variability of runoff estimated with an offline SURFEX simulation over the 1948–2010 period using the global atmospheric forcing from Princeton University (PGF; Sheffield et al., 2006).
In CNRM-ESM1, the marine biophysical feedback is induced by changes in the penetration of downward irradiance in response to marine biota chlorophyll concentration. This feedback mimics the fact that light absorption in the ocean indeed depends on particle concentration and is spectrally selective (Morel, 1988). The implementation of this mechanism is fully described in Lengaigne et al. (2006, 2009) for an ocean forced configuration and Mignot et al. (2013) for a current ocean coupled configuration. It is derived from an accurate 61 spectral band formulation proposed by Morel (1988) using three large wavebands: blue (400–500 nm), green (500–600 nm) and red (600–700 nm). These three bands correspond to the spectral domain of maximum absorption for chlorophyll. The chlorophyll-dependent attenuation coefficients depend on the three-dimensional chlorophyll field predicted by PISCES. They are computed at each time step from a power-law relationship fitting to the coefficients computed from the full spectral model of Morel et al. (1988). This biophysical feedback represents a major evolution from the ocean component used in Voldoire et al. (2013) and Séférian et al. (2013).
The CMIP5 specification requires each model to reach its equilibrium state before kicking off formal simulations, especially for long-term control experiments. To obtain the initial conditions for CNRM-ESM1 preindustrial steady state at year 1850, we first initialize the various physical and biogeochemical components of the model as described below and perform a 400-year-long spin-up simulation using CNRM-ESM1 with all 1850 external forcings (Taylor et al., 2009).
Initialization of the physical components of CNRM-ESM1 relies on previous model outputs from CNRM-CM5.1. This latter model was first initialized from World Ocean Atlas 2005 observations for salinity and temperature (Antonov et al., 2006; Locarnini et al., 2006) and spun up for 200 years. The 801st year of the centennial-long CMIP5 preindustrial run from CNRM-CM5.1 was employed as initial condition for CNRM-ESM1 preindustrial state.
Marine biogeochemical reservoirs were initialized from fields of a previous preindustrial simulation of CNRM-CM5.1 coupled to PISCES. In this previous simulation, PISCES state variables were initialized from World Ocean Atlas 1993 observations for nitrate, phosphate, silicate and oxygen (Levitus et al., 1993) and the Global Ocean Data Analysis Project (Key et al., 2004) for alkalinity and preindustrial dissolved inorganic carbon (DIC). From this initialization, this intermediate version of the ESM was integrated online for 1100 years.
Land biogeochemical reservoirs were initialized from zero and spun up using an acceleration approach for soil carbon and wood during the first century of the spin-up simulation. This approach consists in updating the wood growth, the litter and soil biogeochemistry modules several times per time step with constant incoming carbon fluxes and physical conditions allowing for the various reservoirs of carbon to fill up much faster. As a result of this approach, soil carbon and wood reservoirs were respectively spun up for 21 800 and 1200 years.
Finally, both physical and carbon cycle components of CNRM-ESM1 benefit from an physical adjustment under 1850 preindustrial control conditions for 400 years. Section 4.1 describes the residual drifts of the model at quasi-equilibrium state.
Following CMIP5 specifications (Taylor et al., 2009), CNRM-ESM1 has performed several CMIP5 long-term core experiments and part of the tier-1 experiments.
The preindustrial control simulation,
The 20th century experiment,
Note there is no land-cover change related to anthropogenic land use in the abovementioned simulations. The fraction of vegetal cover is set to the present-day state using the in-house ECOCLIMAP database (Masson et al., 2013a). Therefore, changes in physical and biogeochemical properties of the vegetation due to actual land-cover changes are excluded by design.
To illustrate the stability of CNRM-ESM1 at the end of the spin-up simulation, we show the global average values of a few variables during the 250 years of the piControl simulation (Fig. 2) and their drifts (Table 1).
Time series of various climate indices along the 250-year-long
control simulation.
Drift in climate indices used to evaluate the equilibrium of
CNRM-ESM1's physical and biogeochemical components. The drifts are computed
over the 250-year-long preindustrial simulation of CNRM-ESM1 for the top of
the atmosphere net radiative balance (TOA), the net surface heat flux (NSF),
the near-surface temperature (
In terms of energy balance, the global mean top-of-atmosphere (TOA) net
radiative balance is about 3.57
In terms of global-scale climate indices, the global mean surface
temperature (
With regard to the simulated global carbon cycle, Fig. 2e shows that the
natural carbon cycle is stable over the piControl simulation with
terrestrial and oceanic carbon fluxes of 0.75
In the following, we focus on the physical drivers of the global carbon
cycle. From a land perspective, surface temperature (
Biases in simulated near-surface temperature (
Compared to the CRUTV4 data set (Harris et al., 2013) over the period
1986–2005, CNRM-ESM1 displays a global annual-averaged bias of
Biases in simulated precipitation (PR) compared to the GPCP
observations (Adler et al., 2003) averaged over 1986–2005. Winter
Figure 4 shows the regional structure of the PR bias of CNRM-ESM1 with
respect to the Global Precipitation Climatology Project (GPCP) observations
(Adler et al., 2003). Over continents, CNRM-ESM1 slightly underestimates the
amount of the seasonal PR except over Asia, the western coast of America and
Australia. The major regional bias in seasonal PR is found over Amazonia,
where PR is underestimated by 2 and 5 mm day
Biases in simulated photosynthetically available radiation (PAR)
compared to the Surface Radiation Budget (SRB) satellite-derived observations (Pinker and Laszlo, 1992)
averaged over 1986–2005. Winter
Annual bias patterns of simulated temperature
Compared to Surface Radiation Budget (SRB) satellite-derived observations (Pinker and Laszlo, 1992),
CNRM-ESM1 overestimates the PAR globally (Fig. 5). Major biases are found
over continents except for some regions in the tropics. The magnitude of the
seasonal biases is weaker in Northern Hemisphere winter than in summer when
regional biases reach up to 20–30 W m
From an oceanic perspective, temperature is as important as over land
surface because it sets the marine biota's growth rate, playing a large role
in the biological-mediated processes (e.g., export, soft tissue pump). In
addition, both temperature (
Compared to WOA2013 data products (Levitus et al., 2013), CNRM-ESM1 realistically simulates both the mean annual sea surface temperature and sea surface salinity, both in terms of amplitude and spatial distribution, as shown in Fig. 6a and b. Moderate positive biases in sea surface temperature and sea surface salinity are found in the Southern Ocean and in the eastern boundary upwelling systems. Strong biases in sea surface salinity are found in the Labrador and Arctic seas. While most of these biases are related to an overestimated atmospheric surface heating, biases in the Labrador Sea and in the Arctic are essentially due to erroneous representation of the mixed-layer depth and the Arctic sea-ice cover. These points will be further detailed below.
Composite of yearly extremum of mixed-layer depth over 1986–2005.
Left panels represent the maximum mixed-layer depth (MLD
At depth, the vertical structures in simulated
As mentioned above, an accurate representation of spatial and temporal MLD
is essential for numerous ocean biogeochemical processes. For example,
winter mixing entrains carbon- and nutrient-rich deep waters to the surface,
which play an important role in the transfer of CO
Sea-ice cover (SIC) as simulated by CNRM-ESM1 averaged over 1986–2005. Top panels represent composite of September sea-ice cover, while bottom panels are for March. Iso-15 % of SIC serves as comparison between model results and NSIDC observations (Cavalieri et al., 1996) averaged over 1986–2005; model results and observations are indicated with dashed and solid black lines, respectively.
Compared to the observation-derived estimates, CNRM-ESM1 captures the main
regional pattern of MLD
Taylor diagrams showing the correspondence between model results and
observations for CNRM-ESM1 and CNRM-CM5.2. Near-surface temperature
(
Similarly to the MLD, SIC is an important driver of the ocean carbon cycle.
It constitutes a physical barrier for exchange of CO
In the Antarctic Ocean, Fig. 8b shows that the spatial structures of SIC biases mirror somehow the model–data mismatch in MLD as shown in Fig. 7b. That is, in austral winter, CNRM-ESM1 underestimates SIC where erroneous open-ocean deep convection zones are located, namely, offshore Wilkes Land in the Indian Ocean sector (Fig. 8b). Conversely, too much sea ice is simulated in the Atlantic Ocean sector. As in CNRM-CM5.1, simulated summer Antarctic SIC is strongly underestimated, with very little sea ice surviving summer melt in the Weddell and Ross seas (Fig. 8d).
Impact of coupling frequency on sea-ice cover (SIC) as simulated by CNRM-ESM1 averaged over 1986–2005. Top panels represent composite of September sea-ice cover, while bottom panels are for March. Iso-15 % of SIC serves as comparison between model results using a 6 h coupling frequency (dashed lines) and those using a 24 h coupling frequency (solid lines).
In the following, we compare the skill of CNRM-ESM1 to the closest version of CNRM-CM5 climate model, called CNRM-CM5.2. Figure 9 summarizes skill-assessment metrics for CNRM-ESM1 and CNRM-CM5.2 in terms of major physical drivers of the global carbon cycle (field maps and patterns of errors are presented in Figs. S2 to S7).
The Taylor diagram for land-surface physical drivers clearly demonstrates that CNRM-ESM1 and CNRM-CM5 display comparable skills except for PR (Fig. 9a). Most of the differences in skills are indeed not significant at a 95 % confidence level; models differ solely in terms of PR for which CNRM-ESM1 produces slightly weaker correlation coefficients.
Over the ocean, Fig. 9b shows further differences between both models. The weakest difference in skill concerns SST for which both models display good agreement with WOA2013. With regard to the MLD, CNRM-ESM1 displays a slightly better agreement than CNRM-CM5.2 with observation-derived MLD (Sallée et al., 2010) in terms of correlation but strongly underestimates the spatial variations of this field. Major differences are noticeable for SSS. CNRM-ESM1's skill is clearly lower than that of CNRM-CM5.2. To investigate this difference, we have computed the skill of PR over the ocean, since CNRM-CM5.2 contributes to the spatiotemporal distribution of the SSS concomitantly to the runoff and the sea-ice seasonal cycle. Skill in PR over the ocean is similar for both models (blue diamonds on Fig. 9b). A similar finding is noticed for simulated runoff (not shown). Therefore, the difference in simulated SSS between the two models can be attributed to the revised water conservation interface and erroneous distribution of sea-ice cover. In addition, changes in coupling frequency (i.e., 24 to 6 h) might be at the origin of differences in skills between the two models since it impacts sea-ice cover (Fig. 10).
From the small differences in skill between the two models, we can assume that the inclusion of the global carbon cycle and the biophysical coupling have not noticeably altered the simulated mean-state climate in CNRM-ESM1 compared to that of CNRM-CM5.2.
Now that the physical drivers of the global carbon cycle have been
evaluated, we assess the ability of CNRM-ESM1 to replicate available modern
observations of the terrestrial carbon cycle. We focus on gross primary
productivity (GPP), vegetation autotrophic respiration (Ra) and soil organic
carbon content (cSoil) that control the net natural fluxes of CO
Annual-mean terrestrial gross primary production (GPP). Values are
given for
Regional and global budget of gross primary production (GPP) and terrestrial ecosystem respiration (TER) as simulated by the CNRM-ESM1 and estimated from the FluxNet-MTE data product. Values in brackets indicate the ratio between the autotrophic respiration (Ra) and TER. The uncertainties for the FluxNet-MTE data product derive from the regional partitioning of global mean uncertainties published in Jung et al. (2011). GPP and TER fluxes are determined from a yearly average over 1986–2005.
To evaluate CNRM-ESM1 GPP, we rely on two streams of data, namely, the
FluxNet-Multi-Tree Ensemble
(FluxNet-MTE, Jung et al., 2011) and the MOD17 satellite-derived observations
(Running et al., 2004). Figure 11 shows that the annual mean GPP as simulated
by CNRM-ESM1 is slightly too strong compared to the observed estimates. The
largest model–data mismatch is found in the tropics between 10
Annual-mean autotrophic respiration (Ra) as estimated from MODIS
over 2000–2013
Despite these biases, the global partitioning between vegetation biomass and
soil carbon is realistic with 596.7 and 2105 Pg C compared to the observed
estimates of 560
Stocks of modern soil organic carbon (cSoil) as estimated from the
FAO/IIASA/ISRIC/ISSCAS/JRC (2012) Harmonized World Soil Database
Table 2 shows that CNRM-ESM1 overestimates globally terrestrial ecosystem
respiration (TER) when compared to the up-scaled measurements of
FluxNet-MTE. In the tropics, simulated TER fluxes are 32 % higher than the
FluxNet-MTE estimates. As mentioned above, this bias is essentially due to
an unrealistic Ra, which amounts to 72 % of TER over the sector in the
model. Table 2 shows that the simulated TER is 126.9 Pg C year
Taylor diagrams showing the correspondence between model results
and observations for CNRM-ESM1 and CNRM-CM5.2 (Séférian et al.,
2013). Climatological distribution over 1986–2005 of simulated oxygen
(O
Annual-mean ocean carbon fluxes (fgCO
Compared to the terrestrial carbon cycle, the ocean carbon cycle has already
been implemented in previous versions of CNRM-CM5 (Séférian et
al., 2013). The modeled marine biogeochemistry components have already
benefited from detailed evaluation against modern observations
(Frölicher et al., 2014; Séférian et al., 2013),
analyses of future projections (Laufkötter et al., 2015) and
sensitivity benchmarking (Schwinger et al., 2014). The major
difference between CNRM-ESM1 and previous versions of CNRM-CM5 including a
marine biogeochemistry module lies in the representation of ocean tracers in
the deep ocean. Figure 14 shows that the representation of oxygen,
phosphate, nitrate and silicate fields was improved in CNRM-ESM1 at depth,
except around 1000 m where the strong flow of NADW tends to alter the
distribution of tracers. Below 1500 m, the tracer distribution is in
reasonable agreement with the observations with correlation coefficients
In terms of carbon cycling into the ocean, Fig. 15 shows the simulated
mean annual sea–air CO
Annual-mean zonal-average anthropogenic carbon
(CO
The storage of anthropogenic CO
In this section, we assess the performance of CNRM-ESM1 in terms of two
ecosystem dynamics parameters, namely, the peak leaf area index (LAI
Composite of yearly maximum of leaf area index (LAI
With regard to LAI
Annual-mean surface chlorophyll concentrations (Chl) as estimated
from SeaWiFS over 1997–2010
With regard to ocean Chl, Fig. 18 shows that CNRM-ESM1 displays a reasonable agreement with satellite-derived observations (O'Reilly et al., 1998). Although regional patterns of Chl concentrations were improved compared to that of CNRM-CM5 (Séférian et al., 2013), major model discrepancies are found in oligotrophic gyres and equatorial upwellings. Biases are more pronounced in the Southern Hemisphere where the model fails to produce very low Chl in the southern Pacific gyres. CNRM-ESM1 also fails at capturing western border high Chl concentrations in relation with the equatorial upwelling. Underestimated Chl concentrations in upwelling systems are essentially due to biases in surface wind forcing as well as to the coarse horizontal and vertical resolution of the ocean model. This model limitation partly explains why Chl concentrations are underestimated in high-latitude oceans. In these domains, high coastal concentrations are captured from satellite sensors but cannot be resolved by the model due to its coarse resolution.
In the present section, we analyze the transient response of various climate
indices to the recent climate forcing from 1901 to 2005. We focus on the
near-surface temperature (
Time series of various climate indices as monitored from available
observations (blue solid line) and as simulated by CNRM-ESM1 (red solid line)
since 1901 with global near-surface temperature
Modern mean-state, interannual variability (IAV) and decadal trends
of various global climate indices: the near-surface temperature (
Figure 19 shows that the transient response of
The recent evolution of LCS and OCS agrees with the range of
observation-based and model-derived estimates (Le
Quéré et al., 2014; Takahashi et al., 2010) with an uptake of
CO
In this article, we evaluate the ability of the Centre National de Recherches Météorologiques Earth system model version 1 (CNRM-ESM1) to reproduce the modern carbon cycle and its prominent physical drivers. CNRM-ESM1 derives from the atmosphere–ocean general circulation model CNRM-CM5 (Voldoire et al., 2013) that has contributed to CMIP5 and to the fifth IPCC assessment report. This model employs the same resolution and components as CNRM-CM5 although it uses updated versions of the atmospheric model (ARPEGE-CLIMAT v6.1), surface scheme (SURFEXv7.3) and sea-ice model (GELATO6) in addition to a 6 h coupling frequency. Several biophysical coupling processes are enabled in CNRM-ESM1 thanks to the terrestrial carbon cycle module ISBA (Gibelin et al., 2008) and the marine biogeochemistry module PISCES (Aumont and Bopp, 2006). They consist of the land biosphere-mediated evapotranspiration feedback and the ocean biota heat-trapping feedbacks.
Skill-score matrix based on
Since an earlier version of CNRM-CM5 including the marine biogeochemistry
module PISCES was distributed and used in several studies
(Frölicher et al., 2014; Laufkötter et al.,
2015; Schwinger et al., 2014; Séférian et al., 2013), the inclusion
of the terrestrial carbon cycle module ISBA constitutes the major
advancement in the CNRM-ESM1 development. Although the ISBA terrestrial
carbon cycle module was developed at CNRM in the 2000s, it had never been
coupled to an atmosphere–ocean model and run for long climate simulations.
Here, we show that ISBA embedded in CNRM-ESM1 reproduces the general pattern
of the vegetation and soil carbon stock over the last decades. Although the
photosynthesis scheme in ISBA differs from the other state-of-the-art
process-based models (e.g., Dalmonech et al., 2014), the model displays
similar behavior. That is, it overestimates both the land–vegetation gross
primary productivity and the terrestrial ecosystem respiration. The
compensation between these two fluxes leads to a correct land carbon sink
over the modern period that agrees with the most up-to-date estimates
(Friedlingstein et al., 2010; Jung et al., 2011; Le
Quéré et al., 2014). The largest model–data mismatch is found in the
tropics where the gross uptake of CO
With regard to the marine biogeochemistry component, CNRM-ESM1 produces results
in terms of biogeochemical variables that are comparable to other IPCC-class
ocean biogeochemical models (Fig. 20). The global distribution of
biogeochemical tracers such as oxygen, nutrients and carbon-related fields
has been improved with respect to an earlier model version presented in
Séférian et al. (2013) (Figs. 14 and S9). This change is
attributed to a stronger northward flow of deep water masses from the
Southern Ocean, which improves the vertical distribution of biogeochemical
tracers. However, the strengthening of the meridional flow of deep water
masses has also distorted the vertical structure of some carbon-related
fields. Indeed, the unrealistic flow of North Atlantic deep water of about
26.1 Sv tends to deplete the stock of anthropogenic carbon storage between
surface and 1200 m (Fig. 16c) and consequently to increase it at depth.
Since biases in anthropogenic carbon storage compensate across the water
column, the simulated anthropogenic carbon storage agrees with 1994
observation-based estimates. With regard to the ocean carbon sink, CNRM-ESM1
simulates a global ocean carbon sink that falls within the lower range of
the combination of observation and model estimates over the recent years
(Le Quéré et al., 2014). This slightly underestimated
carbon sink is attributed to larger outgassing of natural CO
We show that CNRM-ESM1 displays results comparable to those of CNRM-CM5 in spite of the inclusion of the global carbon cycle and various biophysical feedbacks. Simulated near-surface temperature, precipitation, incoming shortwave radiation over continents as well as temperature, salinity and mixed-layer depth over oceans broadly agree with observations or satellite-derived product. Except for the salinity and the mixed-layer depth, CNRM-ESM1 display quite similar skill at simulating physical drivers of the global carbon cycle compared to CNRM-CM5. Such a comparison demonstrates the reliability of this model to produce suitable simulations for future climate change projection and impacts studies.
In addition to preindustrial control and historical simulations discussed in this article, several other simulations were performed with CNRM-ESM1 following both the Coupled Model Intercomparison Project Phase 5 (CMIP5) and the Geoengineering Model Intercomparison Project (GeoMIP) experimental design. The CNRM-ESM1 model outputs (referred as “CNRM-ESM1”) are available for download on Earth System Grid Federation (ESGF) under CMIP5 and GeoMIP projects.
A number of model codes developed at CNRM, or in collaboration with CNRM
scientists, is available as open-source code (see
We thank the two anonymous reviewers for their constructive comments and suggestions of the discussion paper. This work was supported by Météo-France, CNRS and CERFACS. We particularly acknowledge the support of the team in charge of the CNRM-CM climate model. Supercomputing time was provided by the Météo-France/DSI supercomputing center. Data are published thanks to the ESGF and IS-ENES2 projects. Finally, we are grateful to C. Frauen for her kind advice on the nuances of the English language.Edited by: O. Marti