• Open Access

Global carbon budgets simulated by the Beijing Climate Center Climate System Model for the last century

Authors

  • Tongwen Wu,

    Corresponding author
    1. Beijing Climate Center, China Meteorological Administration, Beijing, China
    2. Laboratory for Climate Studies, China Meteorological Administration, Beijing, China
    • Corresponding author: T. Wu, National Climate Center, China Meteorological Administration, 46 Zhongguancun Nandajie, Beijing 100081, China. (twwu@cma.gov.cn)

    Search for more papers by this author
  • Weiping Li,

    1. Beijing Climate Center, China Meteorological Administration, Beijing, China
    2. Laboratory for Climate Studies, China Meteorological Administration, Beijing, China
    Search for more papers by this author
  • Jinjun Ji,

    1. Beijing Climate Center, China Meteorological Administration, Beijing, China
    2. Institute of Atmospheric Physics, Chinese Academy of Sciences, Beijing, China
    Search for more papers by this author
  • Xiaoge Xin,

    1. Beijing Climate Center, China Meteorological Administration, Beijing, China
    2. Laboratory for Climate Studies, China Meteorological Administration, Beijing, China
    Search for more papers by this author
  • Laurent Li,

    1. Beijing Climate Center, China Meteorological Administration, Beijing, China
    2. Laboratoire de Météorologie Dynamique, IPSL, CNRS/UPMC, Paris, France
    Search for more papers by this author
  • Zaizhi Wang,

    1. Beijing Climate Center, China Meteorological Administration, Beijing, China
    2. Laboratory for Climate Studies, China Meteorological Administration, Beijing, China
    Search for more papers by this author
  • Yanwu Zhang,

    1. Beijing Climate Center, China Meteorological Administration, Beijing, China
    2. Laboratory for Climate Studies, China Meteorological Administration, Beijing, China
    Search for more papers by this author
  • Jianglong Li,

    1. Beijing Climate Center, China Meteorological Administration, Beijing, China
    2. Laboratory for Climate Studies, China Meteorological Administration, Beijing, China
    Search for more papers by this author
  • Fang Zhang,

    1. Beijing Climate Center, China Meteorological Administration, Beijing, China
    2. Laboratory for Climate Studies, China Meteorological Administration, Beijing, China
    Search for more papers by this author
  • Min Wei,

    1. National Meteorological Information Center, China Meteorological Administration, Beijing, China
    Search for more papers by this author
  • Xueli Shi,

    1. Beijing Climate Center, China Meteorological Administration, Beijing, China
    2. Laboratory for Climate Studies, China Meteorological Administration, Beijing, China
    Search for more papers by this author
  • Fanghua Wu,

    1. Beijing Climate Center, China Meteorological Administration, Beijing, China
    2. Laboratory for Climate Studies, China Meteorological Administration, Beijing, China
    Search for more papers by this author
  • Li Zhang,

    1. Beijing Climate Center, China Meteorological Administration, Beijing, China
    2. Laboratory for Climate Studies, China Meteorological Administration, Beijing, China
    Search for more papers by this author
  • Min Chu,

    1. Beijing Climate Center, China Meteorological Administration, Beijing, China
    2. Laboratory for Climate Studies, China Meteorological Administration, Beijing, China
    Search for more papers by this author
  • Weihua Jie,

    1. Beijing Climate Center, China Meteorological Administration, Beijing, China
    2. Laboratory for Climate Studies, China Meteorological Administration, Beijing, China
    Search for more papers by this author
  • Yiming Liu,

    1. Beijing Climate Center, China Meteorological Administration, Beijing, China
    2. Laboratory for Climate Studies, China Meteorological Administration, Beijing, China
    Search for more papers by this author
  • Fang Wang,

    1. Beijing Climate Center, China Meteorological Administration, Beijing, China
    2. Laboratory for Climate Studies, China Meteorological Administration, Beijing, China
    Search for more papers by this author
  • Xiangwen Liu,

    1. Beijing Climate Center, China Meteorological Administration, Beijing, China
    2. Laboratory for Climate Studies, China Meteorological Administration, Beijing, China
    Search for more papers by this author
  • Qiaoping Li,

    1. Beijing Climate Center, China Meteorological Administration, Beijing, China
    2. Laboratory for Climate Studies, China Meteorological Administration, Beijing, China
    Search for more papers by this author
  • Min Dong,

    1. Beijing Climate Center, China Meteorological Administration, Beijing, China
    2. Laboratory for Climate Studies, China Meteorological Administration, Beijing, China
    Search for more papers by this author
  • Xiaoyun Liang,

    1. Beijing Climate Center, China Meteorological Administration, Beijing, China
    2. Laboratory for Climate Studies, China Meteorological Administration, Beijing, China
    Search for more papers by this author
  • Yang Gao,

    1. Beijing Climate Center, China Meteorological Administration, Beijing, China
    2. Laboratory for Climate Studies, China Meteorological Administration, Beijing, China
    Search for more papers by this author
  • Jie Zhang

    1. Beijing Climate Center, China Meteorological Administration, Beijing, China
    2. Laboratory for Climate Studies, China Meteorological Administration, Beijing, China
    Search for more papers by this author

Abstract

[1] The paper examines terrestrial and oceanic carbon budgets from preindustrial time to present day in the version of Beijing Climate Center Climate System Model (BCC_CSM1.1) which is a global fully coupled climate-carbon cycle model. Atmospheric CO2 concentration is calculated from a prognostic equation taking into account global anthropogenic CO2 emissions and the interactive CO2 exchanges of land-atmosphere and ocean-atmosphere. When forced by prescribed historical emissions of CO2 from combustion of fossil fuels and land use change, BCC_CSM1.1 can reproduce the trends of observed atmospheric CO2 concentration and global surface air temperature from 1850 to 2005. Simulated interannual variability and long-term trend of global carbon sources and sinks and their spatial patterns generally agree with other model estimates and observations, which shows the following: (1) Both land and ocean in the last century act as net carbon sinks. The ability of carbon uptake by land and ocean is enhanced at the end of last century. (2) Interannual variability of the global atmospheric CO2 concentration is closely correlated with the El Niño-Southern Oscillation cycle, in agreement with observations. (3) Interannual variation of the land-to-atmosphere net carbon flux is positively correlated with surface air temperature while negatively correlated with soil moisture over low and midlatitudes. The relative contribution of soil moisture to the interannual variation of land-atmosphere CO2 exchange is more important than that of air temperature over tropical regions, while surface air temperature is more important than soil moisture over other regions of the globe.

1 Introduction

[2] It has been well documented that human activities are enhancing the greenhouse effect of the atmosphere and altering the global climate [Le Treut et al., 2007]. The increase of global temperature during the last 150 years can be at least partially attributed to the increase of atmospheric CO2 concentration as a consequence of anthropogenic activities including combustion of fossil fuels, cement production, and land use-associated emission [Houghton et al., 2001]. At global scale, however, only about half of the anthropogenic carbon emissions have been stored in the atmosphere, while the remainder has been absorbed by oceans and terrestrial biosphere [Prentice et al., 2001].

[3] Increase of CO2 in the atmosphere and the consequent climate change may modify behaviors of the terrestrial ecosystems and ocean biogeochemistry, which is susceptible to introduce important feedbacks in the earth system. Increasing attention has been paid to this issue in the scientific literature [e.g., Sarmiento and Le Quéré, 1996; Sarmiento et al., 1998; Cox et al., 2000; Cramer et al., 2001; Dufresne et al., 2002; Wigley, 2005; Friedlingstein et al., 2001, 2003, 2006; Matthews et al., 2005; Meehl and Coauthors, 2007; Arora et al., 2009; Gregory et al., 2009; Randerson et al., 2009; Boer and Arora, 2010; Cadule et al., 2010; Roy et al., 2011; Vichi et al., 2011; Zickfeld et al., 2011]. Friedlingstein and Prentice [2010] gave a review of earlier works on carbon-climate feedback and concluded that global warming leads to an additional release of CO2 from the land/ocean system to the atmosphere on timescales ranging from interannual to millennial. Zickfeld et al. [2011] reported that strong nonlinearity exists in carbon-climate feedbacks.

[4] Studies in the past several years also revealed existence of large uncertainties in estimation of terrestrial and oceanic carbon uptake. Bolin and Sukumar [2000] showed that terrestrial carbon uptake was in the range of 0.6 to 3.2 GtC yr−1 (gigatons of carbon per year) for the 1980s and 1.0 to 3.6 GtC yr−1 for the 1990s. These numbers have been revised by House et al. [2003], who reported that terrestrial carbon uptake was in the range of 0.3 to 4.0 GtC yr−1 and 1.6 to 4.8 GtC yr−1 for the 1980s and 1990s, respectively. Le Quéré [2010] presented a new estimate of the present-day global carbon budget and pointed out that about 45% of the total CO2 emitted from fossil fuel burning and land use change stayed in the atmosphere on average during the past decades. They also suggested that the efficiency of carbon sinks could have already decreased in the past decades. Considerable uncertainty exists not only in magnitude but also in locations of the carbon sinks and sources [Friedlingstein et al., 2006; Boer and Arora, 2010]. Cadule et al. [2010] pointed out that uncertainty in the amplitude of climate-carbon feedback is mainly due to uncertainty of the response of the terrestrial biosphere to climate change. Various processes contribute to the uncertainty, including temperature dependence of heterotrophic respiration [Cox et al., 2000; Prentice et al., 2001; Zeng et al., 2004], amount of primary productivity [Matthews et al., 2005], vertical mixing in oceans [Friedlingstein et al., 2003], and cycling of carbon in the living biomass especially in the tropical forest [Cox et al., 2004; Denman et al., 2007].

[5] One obstacle in investigation of the carbon cycle is the difficulty in obtaining direct observational estimates of carbon uptake. Climate models are, at present, the most advanced tool to investigate carbon budget and to project future climate. The Coupled Carbon Cycle Climate Model Intercomparison Project (C4MIP) was initiated to evaluate uncertainties of climate-carbon feedbacks [Trenberth et al., 2007]. A few works have already been reported from the scientific community [e.g., Randerson et al., 2009; Cadule et al., 2010; Roy et al., 2011]. Recent studies are coordinated in the phase five of the Coupled Model Intercomparison Project (CMIP5). One main motivation of CMIP5 is to address relevant scientific questions on carbon-climate interactions [Taylor et al., 2009]. The Beijing Climate Center Climate System Model version BCC_CSM1.1 is one of the comprehensive carbon-climate models joining the CMIP5 efforts to support the Intergovernmental Panel on Climate Change (IPCC) AR5.

[6] An accurate simulation of the 20th century global carbon cycle is a prerequisite to reliably project future climate. The purpose of this paper is to evaluate BCC_CSM in reproducing the global carbon cycle from 1850 to 2005 and to quantify the simulated carbon sources and sinks from interannual variability to long-term trend. We also provide some discussions on results of BCC_CSM compared to those of other models.

[7] A relevant description on the model and experiments is firstly presented in section 2. Regional and global characteristics of the simulated land-atmosphere and ocean-atmosphere carbon exchange compared to estimates from other studies are analyzed in section 3.

2 Model Description and Experiments

[8] The model used in this work is the version 1.1 of the Beijing Climate Center Climate System Model (BCC_CSM1.1) developed at the Beijing Climate Center (BCC), China Meteorological Administration (CMA). BCC_CSM1.1 is a fully coupled global climate-carbon model including interactive vegetation and global carbon cycle, in which the atmospheric component BCC Atmospheric General Model version 2.1 (BCC_AGCM2.1), ocean component Modular Ocean Model version 4 (MOM4)-L40, land component BCC Atmosphere and Vegetation Interaction Model version 1.0 (BCC_AVIM1.0), and sea ice component [sea ice simulator (SIS)] are fully coupled and interact with each other through fluxes of momentum, energy, water, and carbon at their interfaces. Information between the atmosphere and the ocean is exchanged once per simulated day. The exchange of atmospheric carbon with the land biosphere is calculated at each model time step (20 min).

2.1 The Atmospheric Model

[9] The atmospheric component in BCC_CSM1.1 is the Beijing Climate Center Atmospheric General Circulation Model version 2.1 (BCC_AGCM2.1). It is a global spectral model with a horizontal resolution of T42, approximately 2.8125° × 2.8125° transformed grid, and 26 levels in a hybrid sigma/pressure vertical coordinate system with the top level at 2.914 hPa. The dynamical core of the model is described in Wu et al. [2008] and a precedent version, BCC_AGCM2.0, is detailed in Wu et al. [2010]. The governing equations of the model are originated from the Eulerian dynamics in the Community Atmosphere Model (CAM3) [Collins et al., 2004], but substantial changes concerning the governing equations and their resolving technique (use of reference atmospheric temperature and surface pressure) have been implemented in BCC. Most of the physical processes are from CAM3 developed by the National Center for Atmospheric Research (NCAR). A few new schemes are implemented, including parameterizations for the deep cumulus convection, dry adiabatic adjustment, latent heat and sensible heat fluxes over ocean surface, and snow cover fraction [Wu et al., 2010]. BCC-AGCM2.1 is an updated version of BCC_AGCM2.0 with a new deep penetrative convection scheme as described in Wu [2012]. Furthermore, CO2 is a prognostic variable in BCC-AGCM2.1. It is no more a passive tracer, and it is calculated through a budget equation, as a function of global-integrated anthropogenic CO2 emissions, and interactive CO2 fluxes at the interfaces with land and ocean. For the time being, CO2 is homogeneously distributed for the whole atmosphere since we do not take into account its spatial variability due to atmospheric circulation and chemical processes.

[10] The atmospheric CO2 concentration (CO2atm) can be formulated as

display math(1)

where t is the time, Efossil and Elanduse are the CO2 emissions due to fossil fuel consumption and cement manufacture, and land use change (including wood harvest), respectively. The CO2 emission data (the total of Efossil and Elanduse) from the years 1850 to 2005 were CMIP5-recommended (available from http://cmip-pcmdi.llnl.gov/cmip5/) and created by Meinshausen et al. [2011]. Fland and Focean in equation (1) are the carbon exchange rates with land and ocean (positive as a flux to atmosphere), respectively. So, negative Fland and Focean represent the net uptake by the land vegetation-soil system and the ocean, respectively.

2.2 The Land Model

[11] The land model in BCC_CSM1.1 is the Beijing Climate Center Atmosphere and Vegetation Interaction Model version 1.0 (BCC_AVIM1.0). It is a comprehensive land surface model and can be coupled into the BCC_CSM1.1 to simulate land surface biogeophysical and plant ecophysiological processes. There are exchanges of energy, water, and carbon between land surface and the atmosphere.

[12] BCC_AVIM1.0 is originated from the Atmosphere and Vegetation Interaction Model (AVIM) [Ji, 1995; Lu and Ji, 2006; Ji et al. 2008] and includes three submodules: biogeophysical, ecophysiological, and soil carbon-nitrogen dynamical modules. A modified biogeophysical framework with 10 layer soil and at most five layer snow is almost the same as that in the NCAR Community Land Model version 3.0 (CLM3) [Oleson et al., 2004]. Since the snow cover fraction (SCF) is underestimated in Bermuda Atlantic Time-series Study [Yang et al., 1997] and CLM3, we adopted a scheme of SCF from the work of Yang et al. [1997] and Roesch et al. [2001] which takes into account the influence of subgrid topography variability on SCF.

display math(2)

where fsno represents SCF, hsno snow depth (m), z0g roughness length of bare soil, Sn snow water equivalence (mm), σz the spatial variance of topography (m) in the grid cell, and ε a minute constant (0.0001).

[13] The terrestrial carbon cycle in BCC_AVIM1.0 operates through a series of biochemical and physiological processes on photosynthesis and respiration of vegetation. There is a seasonally varying allocation of carbohydrate to leaves, stem, and root tissues in function of the prognostic leaf area index. Our model also takes into account carbon loss due to turnover and mortality of vegetation, and CO2 release into atmosphere through soil respiration. The vegetation litter to the ground surface and into the soil is divided into eight terrestrial carbon pools (surface structural, surface metabolic, surface microbial, soil structural, soil metabolic, soil microbial, slow, and passive carbon pools) according to the timescale of the decomposition of carbon in each pool and transfers between different pools on the basis of the CENTURY ecosystem model [Parton et al., 1988] and a model of carbon exchange between vegetation, soil, and the atmosphere (CEVSA) [Cao and Woodward, 1998].

[14] Carbon uptake by the terrestrial vegetation-soil system is formulated as

display math(3)
display math(4)

where Cveg and Csoil are the carbon stored in terrestrial vegetation and soil, respectively. GPP is the terrestrial vegetation gross primary productivity calculated in the biogeophysical module, Lveg is the litterfall from vegetation, Rveg and Rsoil are the plant autotrophic respiration and soil heterotrophic respiration, respectively. Details of the scheme are described in Ji et al. [2008]. The net primary productivity (NPP), which represents the net amount of CO2 taken up by vegetation, is calculated as

display math(5)

[15] From equations ((3))–((4)), the net ecosystem CO2 flux can be derived as

display math(6)

[16] Positive value of NEP indicates net carbon uptake by the terrestrial ecosystem from the atmosphere. There is fland = − NEP in equation (1).

[17] The integration time step is 20 min for the photosynthesis of vegetation, 24 h for the biomass accumulation and phenological variation as well as soil carbon decomposition processes. The vegetated surfaces are divided into 15 plant functional types (PFTs) including natural vegetation and crop as the present situation. In BCC_AVIM1.0, a grid cell contains up to four PFTs which is similar to CLM3. The composition and abundance of PFTs within a grid cell are time-invariant and prescribed from 1 km satellite data [Bonan et al., 2002].

[18] In this study, only the land use-associated CO2 emissions are involved in terms of Elanduse in equation (1). The actual land-cover changes as boundary conditions of the BCC_AVIM1.0 are not involved. Fractional land use patterns are fixed in BCC-CSM1.1 as those in 1850, therefore changes in physical and biogeochemical properties of the vegetation following actual land-cover changes are neglected.

2.3 The Oceanic Model

[19] MOM4_L40 is a global oceanic general circulation model (OGCM) with a tripolar grid of Murray [1996]. The horizontal resolution is 1° longitude by 1/3° latitude between 30°S and 30°N ranged to 1° latitude at 60°S and 60°N and nominally 1° polarward with tripolar coordinates to resolve the arctic. There are 40 z-levels in the vertical. The two northern poles of the curvilinear grid are shifted to land areas over North America and Eurasia, respectively. The first 20 levels are placed between surface and 200 m depth of the upper ocean. MOM4-L40 is originated from the Z-coordinate Modular Ocean Model version 4 (MOM4) developed by the Geophysical Fluid Dynamics Laboratory (GFDL). It adopts some mature parameterization schemes in MOM4 [Griffies et al., 2005], including Sweby's tracer-based third-order advection scheme, isopycnal tracer mixing and diffusion scheme, Laplace horizontal friction scheme, KPP vertical mixing scheme, complete convection scheme, overflow scheme of topographic processing of sea bottom boundary/steep slopes, and shortwave penetration schemes based on spatial distribution of chlorophyll concentration. The biogeochemistry module to simulate the ocean carbon cycle in MOM4_L40 is based on the protocols from the Ocean Carbon Cycle Model Intercomparison Project–Phase 2 (OCMIP2, http://www.ipsl.jussieu.fr/OCMIP/ phase2/).

[20] The OCMIP biogeochemistry module parameterizes the process of marine biology in terms of geochemical fluxes without explicit representation of the marine ecosystem and food web processes. It includes five prognostic variables: phosphate (PO4), dissolved organic phosphorus (DOP), dissolved oxygen (O2), dissolved inorganic carbon (DIC), and alkalinity (Alk). In OCMIP, Focean in equation (1) is calculated as

display math(7)

[21] In which,

display math(8)

and

display math(9)

where Kw is the gas transfer velocity [Wanninkhof, 1992], Fice is the fractional sea ice cover, u is the wind speed near surface, Sc is the Schmidt's number for CO2 [Wanninkhof, 1992]. αc is the CO2 solubility for water vapor saturated air, pCO2 atm is the partial pressure of CO2 in dry air, psea and patm are the pressure at sea surface level and at the lowest layer of the atmospheric model, respectively. In equation (9), CO2 surf is the surface aqueous CO2 concentration varying in function of the oceanic surface DIC, Alk, temperature, and salinity. The DIC depends on the air-sea exchange of CO2 and is given by

display math(10)

where L(DIC) represents effects due to dynamic processes of advection and diffusion. Jb(DIC) and Jg(DIC) are the source/sink terms due to biological process and air-sea exchange of CO2, respectively. Jg(DIC) can be written as

display math(11)

where dz is the top layer thickness of MOM-L40.

[22] Carbon cycle processes in MOM4-L40 are kept identical to those in OCMIP, except for parameterizing the export of organic matter from surface waters to deep oceans. This is generally known as “export production” (EP for short, also called “new production”) and is a very important process in determining the carbon cycle. In MOM4_L40, the EP is simulated with a prognostic method following Yamanaka and Tajika [1996], while the nutrient restoring approach is used in the original OCMIP. That is, the EP in MOM4_L40 is parameterized as a function of phosphate concentration,

display math(12)

where r is a proportional factor called “bio-production efficiency” and is set to 0.8 yr−1 in this work, and Lf is the light factor related to strength of the incident solar radiation [Bacastow and Maier-Reimer, 1990].

2.4 The Sea Ice Model

[23] The sea ice component of BCC_CSM1.1 is the GFDL Sea Ice Simulator (SIS). SIS is a global dynamical-thermodynamical sea ice model, where the elastic-viscous-plastic technique [Hunke and Dukowicz, 1997] is used to calculate ice internal stresses and the thermodynamics is a modified Semtner scheme from Winton [2000]. SIS has the same horizontal resolution as MOM4-L40 and three layers in the vertical, including one layer of snow cover and two layers of equally sized sea ice. In each model grid, five categories of sea ice are considered, according to the thickness of sea ice. It also takes into account the mutual transformation from one category to another under thermodynamic conditions. SIS calculates concentration, thickness, temperature, salinity of sea ice, and motions of ice sheet. More details can be found in Winton [2000].

2.5 Experiments

[24] The preindustrial climate state of BCC_CSM1.1 is obtained from a 300 year control simulation following the requirement of CMIP5. At first, the initial state of each component of BCC_CSM1.1 is obtained through individual “spinup” runs. The atmospheric model is integrated for 50 years forced by fixed preindustrial forcing of the year 1850, including global reconstructed SST and sea ice concentration. The land model BCC_AVIM1.0 is forced with a fixed atmospheric CO2 concentration of 285 ppmv in 1850 and the atmospheric forcing of 1950–2000 NCEP reanalyses data in an offline way to yield a first estimate of equilibrium states for the land carbon pools. The ocean component is integrated alone for 200 years using the surface forcings from the above mentioned 50 year atmospheric integration with an atmospheric CO2 content fixed at the value in 1850 for the OCMIP module of MOM4-L40.

[25] All components of BCC_CSM1.1 are finally fully coupled together and integrated, initiated from preindustrial conditions of the year 1850, for 100 years as “spinup,” and then 300 years control run as required by the CMIP5 [Taylor et al., 2009] to approach its quasi-equilibrium state. The globally averaged surface temperature is about 287 K (Figure 1a). The atmospheric CO2 concentration is about 286 ppmv (Figure 1b), very close to the observed estimation of 285 ppmv for 1850.

Figure 1.

Time series of globally averaged annual mean of (a) surface air temperature (K), (b) CO2 concentration (ppmv), (c) land CO2 flux (GtC yr−1), (d) oceanic CO2 flux (GtC yr−1), and (e) global CO2 flux (GtC yr−1) from the 300 year preindustrial control experiment. CO2 fluxes are accounted positive upward.

[26] As indicated in equation (1), the atmospheric CO2 interacts with the carbon budgets in global land and oceans. The inhomogeneous distributions of global vegetation, precipitation, and temperature cause seasonal to interannual variations of regional land carbon exchange with atmosphere, then impact on change in atmospheric CO2 concentration. As shown in Figure 1, the atmospheric CO2 concentration is in the range of 283 to 289 ppmv. There exist evident interannual variations of about −2.0 ~ 2.0 GtC yr−1 carbon exchange between atmosphere and land and −0.2 ~ 0.4 GtC yr−1 between atmosphere and ocean. If we compare Figure 1c with other model simulations such as a similar 1000 year preindustrial control experiment using the NCAR CSM1.4-carbon model in Doney et al. [2006], then the magnitude of variability for the global annual land CO2 flux from BCC_CSM1.1 control run is slightly larger.

[27] In the 300 year preindustrial control experiment, the entire ocean is always, on the average, a weak carbon source, and there is a small amount of net carbon flux from ocean to atmosphere, estimated at about 0.159–0.174 GtC yr−1 (Table 1) averaged for every 50 years. At the same time, the entire land acts as a net carbon sink at about 0.042–0.173 GtC yr−1 for every 50 years (Table 1). In the final 50 years of the 300 year preindustrial experiment, they are approximately balanced, only a small CO2 flux (0.001 GtC yr−1) to the atmosphere, on average (Table 1). Therefore, the BCC_CSM1.1 approaches its quasi-equilibrium state under the preindustrial condition without any anthropogenic CO2 emission. We notice that the recent works of Liu et al. [2010] and Ludwig et al. [2011] revealed a carbon transfer of 0.24–0.60 GtC yr−1 to the sea via continental rivers. This may have some influence on the carbon budget in the ocean, and even possibly to partly offset the BCC_CSM1.1 simulated net carbon source in the ocean, but it is not included in the BCC_CSM1.1.

Table 1. Globally Averaged Annual Means of CO2 Concentration (ppmv), Land CO2 Flux (GtC yr−1), Oceanic CO2 Flux (GtC yr−1), and Global CO2 Flux (GtC yr−1) Averaged for Every 50 Years in the 300 Year Preindustrial Control Experimenta
YearsCO2 Concentration (ppmv)Land CO2 Flux (GtC yr−1)Oceanic CO2 Flux (GtC yr−1)Global CO2 Flux (GtC yr−1)
  1. a

    CO2 fluxes are accounted positive upward.

1–50285.566−0.1680.167−0.001
51–100285.495−0.0420.1590.117
101–150285.474−0.1260.1700.044
151–200287.077−0.0740.1100.036
201–250286.823−0.1520.1740.022
251–300286.487−0.1730.1720.001

[28] The steady state of BCC_CSM1.1 after the preindustrial control experiment is used as the initial condition for the historical experiment. The model is integrated from 1850 to 2005 with the CMIP5 recommended prescribed historical CO2 emissions including that from fossil fuel burning and other associated with historical land use changes [Meinshausen et al., 2011]. The simulation also includes all other prescribed historical forcing such as insolation, orbital forcing, tropospheric and stratospheric sulfates, volcanic aerosol, ozone, and non-CO2 greenhouse gases like CH4, N2O, CFC-11, and CFC-12. All of these forcing data are downloaded from CMIP5 website (http://cmip-pcmdi.llnl.gov/cmip5/data_portal.html).

3 Results

3.1 Performance in Simulating the Global Warming in the Last Century

[29] The ability to reproduce the global warming in the 20th century is a key point for a climate-carbon cycle model. According to the CMIP5 protocol, as shown in Figure 2a, the globally integrated anthropogenic CO2 emission is 0.5 GtC yr−1 in year 1850. It persistently increases year by year thereafter, mainly due to increase of CO2 release from fossil fuel burning and cement production. The total anthropogenic CO2 emission reaches about 2 GtC yr−1 around 1950, and its increase is accelerated afterwards. Up to year 2000, the global total anthropogenic emission reaches about 8 GtC yr−1.

Figure 2.

Time series of (a) the CMIP5-recommended global annual anthropogenic CO2 emissions (thick solid line) due to fossil fuel combustion and cement production (dash line) and other anthropogenic activities (thin solid line), (b) the annual CO2 concentration, and (c) the global mean surface air temperature anomalies from the historical experiment (widen solid line) compared to observations (thin solid line). The units are (a) GtC yr−1, (b) ppmv, and (c) K, respectively. The observation data in Figures 2b and 2c are the CMIP5-recommended global CO2 observation data set and the HadCRUT3 data set [Brohan et al., 2006], respectively.

[30] Under the forcing of anthropogenic carbon emissions (Figure 2a), BCC_CSM1.1 can well reproduce the atmospheric CO2 concentration and its time evolution. As shown in Figure 2b, the simulated atmospheric CO2 increase is in close agreement with CMIP5 recommended values (based on observations) from 1850 to 2005. The simulated atmospheric CO2 concentration started from 286 ppmv in 1850 and increased to 385 ppmv by 2005 (Figure 2b), and the discrepancy between the simulation and observation is smaller than 10 ppmv. There is a systematic underestimate of CO2 in BCC_CSM1.1 from 1880 to 1950, which also exists in other model studies [e.g., Matthews et al., 2005; Eby et al., 2009; Arora et al., 2009; Murakami et al., 2010]. In the latter half of the 20th century, the simulated CO2 concentration in BCC_CSM1.1 is, however, overestimated by 5 ~ 10 ppmv relative to the observation.

[31] With the increase of atmospheric CO2 concentration and all other historical forcing data including solar activity, aerosol, ozone, and non-CO2 greenhouse gases, the global warming during the 20th century in BCC_CSM1.1 is well simulated (Figure 2c). Globally averaged surface air temperature (SAT) from 1850 to 2005 displays similar time evolution as the observation from HadCRUT3 SAT data set [Brohan et al., 2006], with a well-reproduced long-term increasing trend of SAT. Our simulated 0.7°C increase during 1850–1995 is in good agreement with the IPCC AR4 estimate of 0.6°C ± 0.2°C [Trenberth et al., 2007]. But in the last period from 1996 to 2005, the global SAT experiences a jump of 0.6°C in the model and is obviously warmer than in the observation.

[32] It is noted that the simulated SAT in some time periods such as in the 1880s, 1890s, 1900s, 1960s, and 1990s has an evident global cooling shock. These are coincident to several significant volcanic eruptions such as Krakatoa (in 1883), Pelee (in 1902), West Indies Agung (in 1963), and Mount Pinatubo (in 1991). Each of these volcanic eruptions may lead to a significantly enhanced stratospheric aerosol optical depth (available from http://data.giss.nasa.gov/modelforce/strataer/). As shown in Figure 2c, the global surface air temperature may decrease by up to 0.4°C within 1 to 2 years after a volcanic eruption. This volcano aerosol response is slightly stronger than the ensemble simulation of other CMIP5 models (not shown).

3.2 Variation of Global Carbon Budget

[33] Carbon uptake by global land and ocean plays an important role in slowing down the carbon accumulation in the atmosphere. Figure 3 presents CO2 fluxes simulated by BCC_CSM1.1 for global land and ocean, respectively. Negative values indicate fluxes from atmosphere to land or ocean. Along with the increase of anthropogenic CO2 emission to the atmosphere (Figure 2a) especially after 1950, there is an obvious negative tendency in the second half of the 20th century for both land and ocean CO2 fluxes (Figure 3). This decreasing slope with time indicates that the ability of carbon uptake by land and ocean are both enhanced.

Figure 3.

Time series of the simulated annual mean CO2 flux averaged for the global ocean (thick line) and global land (thin line). The units are GtC yr−1, accounted positive upward.

[34] As a result of carbon absorption by land and ocean, part of CO2 released by human activity stays in the atmosphere. As listed in Table 2, there are 249 GtC of CO2 release from fossil fuel burning and 146 GtC of other CO2 emission to the atmosphere from 1850 to 1998. During the same period, there are 136 and 77 GtC absorbed by the global land and ocean, respectively. Therefore, approximately 46% of the total 395 (i.e., 183) GtC of anthropogenic CO2 emissions still remains in the atmosphere in BCC_CSM1.1 simulation. This leads to increase in model atmospheric CO2 from 285 ppmv in 1850 to 369 ppmv in 1998.

Table 2. Global Carbon Budgets From BCC_CSM1.1 and Other Estimates for the 1980s, 1990s, and 1850–1998.
  1980s (GtC yr−1)1990s (GtC yr−1)1850–1998 (GtC)
  1. a

    From Bolin and Sukumar [2000];

  2. b

    From Plattner et al. [2002];

  3. c

    From Houghton [2003];

  4. d

    From Matthews et al. [2005];

  5. e

    IPCC TAR [Prentice et al., 2001];

  6. f

    Roedenbeck et al. [2003];

  7. g

    Raddatz et al. [2007];

  8. h

    IPCC AR4 [Denman et al., 2007];

  9. i

    Eby et al. [2009];

  10. j

    Estimated from CMIP5-recommended CO2 observation data in this work.

Fossil fuel, cement, and biofuel emissionsForcing data5.26.0249
Land use emissionsForcing data1.51.6146
Atmospheric increaseBCC_CSM1.13.13.6182
 Other estimates3.3 ± 0.2a3.3 ± 0.2a176 ± 10a
3.1d3.4d170d
3.3i3.7i172j
3.3 ± 0.1h3.2 ± 0.1h 
3.3j3.2j 
Ocean uptakeBCC_CSM1.11.61.877
 Other estimates1.7 ± 0.6b2.4 ± 0.7b120 ± 50b
2.2d2.6d132d
1.8i2.2i 
1.9 ± 0.6e1.7 ± 0.5e 
1.8 ± 0.3f2.3 ± 0.2f 
1.96 ± 0.07g2.24 ± 0.07g 
1.8 ± 0.8h2.2 ± 0.4h 
Land uptakeBCC_CSM1.12.02.2136
 Other estimates2.4 ± 1.1c2.9 ± 1.1c126 ± 80c
2.3d2.7d135d
2.2i2.6i 
2.2 ± 0.9e3.0 ± 1.1e 
2.2 ± 0.9f2.6 ± 0.9f 
2.18 ± 0.25g2.01 ± 0.25g 
1.7 (3.4 to -0.2)h2.6 (4.3 to 0.9)h 

[35] It is noted that the averaged ocean carbon uptakes during the 1980s and 1990s are 1.7 and 1.8 GtC yr−1, respectively. They are slightly smaller than most of the previous estimates [Plattner et al., 2002; Matthews et al., 2005; Prentice et al., 2001; Roedenbeck et al., 2003; Raddatz et al., 2007; Meehl et al., 2007; Eby et al., 2009]. However, McNeil et al. [2003] pointed out that most ocean general circulation models overestimated their anthropogenic CO2 uptake over the past 2 decades. With a technique based on global chlorofluorocarbon data, the ocean net uptake was estimated 1.6 GtC yr−1 for the 1980s and 2.0 ± 0.4 GtC yr−1 for the 1990s, respectively. BCC_CSM1.1 is close to these results in McNeil et al. [2003]. Note that a consensus result for the 1990s given in IPCC-AR4 WG1 Ch7 [Denman et al., 2007] is 2.2 GtC yr−1.

[36] The total land CO2 uptake simulated by BCC_CSM1.1 is, on average, 2.0 and 2.2 GtC yr−1 during the 1980s and 1990s, respectively, within the previous estimated uncertainty range [Houghton, 2003; Matthews et al., 2005; Prentice et al., 2001; Roedenbeck et al., 2003; Raddatz et al., 2007; Denman et al., 2007; Eby et al., 2009].

[37] The carbon uptake by land is, to a large extent, determined by vegetation and soil carbon storages. Figure 4 shows temporal evolution of the global vegetation biomass, soil carbon storage, GPP, and NPP simulated by BCC_CSM1.1. From 1850 to 1994, the global carbon storage in vegetation increases from 410 to 470 GtC and storage in soil increases from 990 to 1046 GtC. There is a total of 116 GtC increase of carbon storage in the global land within about 150 years. It is a little larger than the estimate of 101 GtC increase during the period of 1800 to 1994 from IPCC AR4 [Denman et al., 2007] but smaller than the 128 GtC simulated by Eby et al. [2009]. Recent studies using satellite remote sensing data indicated that the net primary production (NPP) has increased globally over the past 2 decades [Nemani et al., 2003; Potter et al., 2003]. As shown in Figures 4c and 4d, the simulated global annual GPP and NPP through the 20th century increase from 105 to 130 GtC and from 55 to 65 GtC, respectively. The annual mean global GPP and NPP averaged for the period of 1980 to 2000 are about 125 and 64 GtC, respectively, which are close to other observation or model-based estimates of the global GPP (125 ~ 139.7 GtC) [Zhao et al., 2006; Arora et al., 2009] and NPP (56.6 ~ 71.1 GtC) [Running et al., 2004; Piao et al., 2009a; Arora et al., 2009]. However, the BCC model estimate of GPP is significantly lower than the recent observational estimates of 150 ~ 175 pgC [Welp et al., 2011] and 127 ~ 166 pgC from atmospheric CO2 data assimilation efforts [Koffi et al., 2012]. We also note that the simulated soil carbon stock as shown in Figure 4b is smaller than most of other estimates of 1220 ~ 2200 pgC [e.g., Sombroek et al., 1993; Kimble et al., 1990; Batjes, 1996]. The underestimate is possibly due to the small stock of slowly evolved passive soil carbon pool in BCC-CSM. In addition, a large amount of carbon captured in permafrost regions is not considered in BCC-CSM1.1.

Figure 4.

Time series of global annual mean of (a) vegetation biomass, (b) soil carbon, (c) gross primary production, and (d) net primary production from the historical experiment. The units are GtC in Figures 4a and 4b, and GtC yr−1 in Figures 4c and 4d, respectively.

3.3 Geographical Distribution

[38] Figure 5 shows the geographic distribution of the annual mean CO2 flux averaged for the last 30 years in the preindustrial experiment and 30 years (1971–2000) in the historical experiment. Regions of major carbon sinks or sources in the 20th century (Figure 5b) are roughly same as those in the preindustrial period (Figure 5a). Eastern U.S., China, and Europe are simulated as regions of significant carbon sink and are consistent with recent estimates from observations in Piao et al. [2009b]. Simulated large carbon sources to the atmosphere are located in central to western Australia, and in the central Amazon. Referring to the natural condition of the preindustrial (Figure 5a), large increases of land carbon uptake in the last 30 years of the 20th century are distributed in central Australia, coastal regions in South America continent, tropical regions, China, eastern coasts of U.S., and high-latitude regions in Russia (Figure 5c).

Figure 5.

Geographical distribution of annual mean CO2 net flux (accounted positive upward, from land to atmosphere) for (a) the last 30 years in the preindustrial experiment, (b) the present climate (1971–2000) in the historical experiment, and (c) the difference between Figures 5b and 5a. The units are g C m−2 yr−1.

[39] Raddatz et al. [2007] showed that tropical regions play a key role in controlling the global carbon-climate positive feedback with simulations from C4MIP. The Amazon contains more than half of the world's tropical rain forests. Some numerical modeling studies [e.g., Prentice and Lloyd 1998] and empirical studies [Malhi et al., 1998] suggested that tropical forests are terrestrial carbon sinks, but recent studies indicated that the sink of tropical Americas is rather weak [Malhi, 2010] or even a source of CO2 when deforestation is taken into consideration [Pan et al., 2011]. As indicated by equation (6) in section 2.2, the carbon exchange between land and atmosphere in the Amazon tropical forest is represented as a small difference of two large terms: the gross fluxes of NPP and the soil respiration, therefore, determination of carbon sink or source is a delicate operation.

[40] Figure 6 presents the spatial distribution of the 1971–2000 averaged annual mean NPP from BCC_CSM historical simulation, as well as results from Moderate Resolution Imaging Spectroradiometer (MODIS) data for 2000–2003 and from the International Geosphere-Biosphere Programme (IGBP) Global NPP Model Intercomparison Data. The annual mean IGBP NPP data were obtained from the Website of International Satellite Land Surface Climatology Project Initiative II (ISLSCP II) which comprises 17 global models with biogeochemistry [Cramer et al., 1999]. We can see that tropical forests in BCC_CSM1.1 contain a large amount of live biomass. Tropical regions see the largest NPP, and boreal forest regions experience the second largest NPP. These results are quite consistent with MODIS and IGBP data. In the Amazon region (50–70°W, 10–20°S), the NPP from BCC model simulation, MODIS, and IGBP estimates accounts for 6.6%, 9.4%, and 8.8% of the global NPP, respectively. Simulated values of NPP in Western Europe, Eastern China, and the southeastern United States are larger than the counterparts in MODIS, but slightly lower than those from IGBP data. The above three regions are all crop fields where a large uncertainty exists in photosynthesis parameterization for crops in climate models. Figure 7 presents scatter plots of NPP as shown in Figure 6 for global grid points from BCC_CSM1.1 and MODIS, and from BCC_CSM1.1 and IGBP. On average, BCC_CSM1.1 is relatively larger than MODIS, especially for high NPP above 0.6 kg C m−2 yr−1. It is closer to IGBP but the maximum of NPP is limited below 1.2 kg C m−2 yr−1.

Figure 6.

Geographical distribution of annual mean net primary production (a) from BCC-CSM1.1 (1971–2000 mean), (b) derived from available MODIS data (2000–2003 mean), and (c) from IGBP data. The units are g C m−2 yr−1.

Figure 7.

Scatter plots comparing annual mean NPP between (a) the BCC-CSM and MODIS, and (b) between BCC-CSM and IGBP. MODIS and IGBP data were firstly interpolated to the mesh of BCC-CSM. Each spatial grid corresponds to a point in the plots. The units are g C m−2 yr−1.

[41] As shown in Figure 6, the maximum model bias comparing with MODIS and IGBP data is located in South America. The simulated NPP in the central part of the Amazon is approximately 0.6 ~ 1.2 kg C m−2 yr−1, which is smaller than the estimations derived from MODIS and IGBP data. It is also smaller than estimates about 1.15 kg C m−2 yr−1 in Senna et al. [2009] and 1.273 ~ 1.350 kg C m−2 yr−1 in Nunes et al. [2012]. This discrepancy is probably caused by insufficient precipitation in the Amazon simulated in BCC_CSM1.1. As presented in Figure 8, the overall patterns of the mean precipitation for December-January-February (DJF) and June-July-August (JJA) from BCC_CSM1.1 resemble their counterparts in observations. The domains covered by larger than 4 mm d−1 are approximately coincident with the observation. The areas with high rainfall in DJF from CPC (Climate Prediction Center) Merged Analysis of Precipitation (CMAP) data over the western parts of the southern tropical Pacific, the southern tropical Indian Ocean, South Africa, and South America, except a weak rain belt over the northern tropical Pacific near the equator, are all well captured by BCC_CSM1.1. The secondary maxima of precipitation over midlatitudes are also reasonably reproduced. In boreal summer (JJA), high rainfall is mainly distributed along the equatorial Pacific and the Asian summer monsoon area and is well reproduced by BCC_CSM1.1. However, remarkable regional negative biases of precipitation from BCC_CSM1.1 exist in the Amazon for both DJF and JJA. This model bias of insufficient precipitation in the Amazon also exists in austral autumn and spring (not shown).

Figure 8.

The 1979–2000 averaged mean precipitation of (left) December-January-February (DJF) and (right) June-July-August (JJA) from (top) BCC_CSM1.1 and (middle) the CMAP observation, and (bottom) their difference. The units are mm d−1.

[42] Figure 9 presents the annual mean air-sea CO2 fluxes averaged over 30 years for the 300 years preindustrial experiment and present-day climate (1971–2000) of the historical experiment. Main spatial patterns for both preindustrial (Figure 9b) and 20th century (Figure 9a) are quite similar to the estimate of Takahashi et al., [2009] (Figure 9d). All the three panels in Figure 9 are characterized by out-gassing of CO2 to the atmosphere from the equatorial Pacific and Atlantic, and by oceanic uptake at high latitudes of the two hemispheres. In the north Pacific and north Atlantic, there are large carbon uptake regions. The intensity and location of those carbon sources or sinks from the observational estimates of Takahashi et al. [2009] are well simulated by BCC_CSM1.1. Comparing with the observational estimate, a large discrepancy is found in the Southern Ocean. Some studies have shown the Southern Ocean south of 40°S to be a large sink for anthropogenic CO2 [Sarmiento and Sundquist, 1992; Sarmiento et al., 1998; Russell et al., 2006; Ito et al., 2010]. But as shown in Figure 9a, the Southern Ocean acts as a natural carbon sink except in a narrow belt of carbon source around 45–60°S in BCC_CSM1.1. Such a region of weak CO2 source also exists in observational estimates of Takahashi et al. [2009].

Figure 9.

Annual mean air-sea CO2 fluxes from (a) the historical simulation in 1971–2000, (b) the preindustrial simulation (last 30 years), and (c) the difference between Figure 9a and 9b, and (d) observational estimates (from Eby et al. [2009]; originated from Takahashi and Coauthors [2009]). Negative values denote ocean uptake, and the units are mol m−2 yr−1.

[43] If we take the natural condition of the preindustrial (Figure 9b) as a reference, the variation of oceanic carbon exchange (Figure 9c) from preindustrial to present-day climate is dominated by strong carbon uptakes in higher latitudes, especially in the North Atlantic and the Southern Ocean.

[44] Figure 10 shows the net annual carbon fluxes to the atmosphere for different latitudinal zones from 1850 to 2005. BCC_CSM1.1 indicates moderate carbon sources in the tropics and carbon sinks in other latitudinal zones. The largest sink is located in northern midlatitudes, which is generally in agreement with other simulation results [e.g., Tans et al., 1990; Gurney et al., 2002] and estimates from CO2 vertical profiles measures [Stephens et al., 2007]. The midlatitude Southern Ocean shows a weak carbon source before 1910 but thereafter begins to weaken and gradually shifts to a weak carbon sink. In most areas of the northern hemisphere, an increase in ability of carbon uptake starts from 1960, and in the tropics, a decrease of the carbon source occurs at almost the same time. But a notable increase in the ability of carbon uptake over the Southern Ocean starts since 1980s.

Figure 10.

Time latitude diagram of zonally averaged annual mean CO2 fluxes from the historical experiment. Units: g C yr−1.

3.4 Interannual to Long-Term Variation for the Recent 50 Years

[45] Table 3 lists the net carbon fluxes over land and ocean for four different latitudinal belts (as defined in Piao et al. [2009a]). In the last 50 years of the 20th century, the tropics still acts as a weak carbon source, and there is a flux of 0.43 GtC yr−1 (to the atmosphere) in the last decade of the 20th century, with main contributions from the tropical oceans (0.53 GtC yr−1). The other three zones show increasing trends of carbon sink. The northern midlatitude land (20°N–50°N) and the Southern Ocean (south of 20°S) are regions of relative strong carbon sink. In the 1990s, the carbon flux is −1.22 GtC yr−1 in northern midlatitude, and −1.29 GtC yr−1 in the Southern Ocean, respectively. The total land carbon uptake is enhanced from 0.41 GtC yr−1 in the 1950s to 2.2 GtC yr−1 in the 1990s, and the global ocean carbon uptake from 0.71 GtC yr−1 in the 1950s to 1.83 GtC yr−1 in the 1990s.

Table 3. Net Carbon Fluxes (Accounted Positive Upward) for Different Latitudinal Zones and Different Time Periodsa
Latitudinal ZonePreindustrial1950s1960s1970s1980s1990s
  1. a

    The units are GtC yr−1. “L,” “O,” and “L + O” denote the carbon flux over land, ocean, and the sum, respectively. In the preindustrial column, numbers are from the last 50 years of the preindustrial control experiment. Negative and positive values show carbon sink and source, respectively.

50°N–90°NL−0.22−0.52−0.57−0.52−0.69−0.71
O−0.17−0.24−0.25−0.25−0.29−0.29
L + O−0.39−0.75−0.82−0.78−0.98−1.00
20°N–50°NL−0.64−0.53−1.42−1.16−1.11−1.22
O−0.45−0.62−0.67−0.70−0.77−0.78
L + O−1.09−1.16−2.09−1.86−1.88−2.01
20°N–20°SL0.580.66−0.130.280.24−0.10
O1.160.920.770.710.620.53
L + O1.741.570.640.990.860.43
20°S–90°SL0.11−0.02−0.090.19−0.41−0.17
O−0.37−0.77−0.83−1.04−1.21−1.29
L + O−0.26−0.78−0.92−0.85−1.62−1.45
Global fluxL−0.17−0.41−2.21−1.21−1.98−2.20
O0.17−0.71−0.97−1.28−1.65−1.83
L + O0.00−1.12−3.18−2.50−3.63−4.04
Anthropogenic emission 0.003.184.355.786.717.63

[46] Previous studies show that interannual variability in the atmospheric CO2 concentration is well correlated with the El Niño–Southern Oscillation (ENSO) cycle [e.g., Bacastow, 1976; Jones et al., 2001]. Figure 11a shows time series of observed annual mean Niño-3 index and natural changes in annual mean atmospheric CO2. Their correlation coefficient from 1959 to 2005 is 0.528, significant at the 99% confidence level. The “observed” globally averaged atmospheric CO2 concentrations are the CMIP5-recommended data and created by Meinshausen et al. [2011]. Natural variability of CO2 concentration is estimated from its annual increment (i.e., the CO2 concentration minus that of the year before) subtracting a linear trend that is considered as the contribution of anthropogenic emissions to atmospheric CO2 variation. The Niño-3 index is the mean SST anomaly in the region 5°N–5°S, 150°W–90°W and downloaded from http://www.metoffice.gov.uk/hadobs/hadisst/data/download.html.

Figure 11.

(a) Time series of the yearly Niño-3 index (°C) (dashed line) and the annual increment of atmospheric CO2 (ppmv) (solid line) derived from the Hadley SST data and CMIP5-recommended CO2 values, and (b) as in Figure 11a but from BCC-CSM1.1.

[47] From BCC_CSM1.1, the natural variability of annual mean atmospheric CO2 concentration is also highly correlated with the model Niño-3 index. Their correlation coefficient is 0.426 and significant at the 95% confidence level. Jones et al. [2001] suggested that this positive correlation between the global atmospheric CO2 and Niño-3 index can be largely attributed to climatic changes over land during El Niño events which lead to decreased gross primary productivity and increased plant and soil respiration.

[48] The close relationship between CO2 flux and Niño3 SST (Figure 11) is mainly attributed to CO2 variation in the tropics. Figure 12 shows a wavelet power spectrum, calculated as in Torrence and Compo [1998] after removing long-term trends of the CO2 flux, for the time series of annual mean CO2 flux from 1850 to 2005 averaged over three zones in the northern midlatitudes, tropics, and southern midlatitudes. These are three regions with remarkable variations for CO2 exchanges of land-atmosphere and ocean-atmosphere (Figure 10). In Figures 12a and 12b, most of the significant power at 95% confidence level is concentrated within the ENSO band of 2–8 years, although there is appreciable power at longer periods. In the tropics, there is a remarkable peak, and the largest amplitude of the wavelet power spectrum is near the cycle of 4 years. But it cannot be discerned in higher latitudes.

Figure 12.

Morlet wavelet power spectrums of the annual mean CO2 flux from 1851 to 2005 averaged for the latitudinal zone of (a) 40°N–60°N, (b) 10°S–10°N, (c) 40°S–60°S. The cross-hatched region denotes the 95% confidence level. The right part in each panel shows integrated power over all scales and times. The left and right axes are the period (in years). The right part of horizontal axis denotes the amplitude of power spectrum (g C yr−1)2.

[49] The late half of the 20th century experiences the most rapid increase of anthropogenic carbon emission to the atmosphere and is also a period of strong carbon uptakes by ocean and land. Figure 13 presents the long-term trends of annual mean net CO2 flux to the atmosphere from 1950 to 2000. Negative (positive) value in Figure 13 means increase (decrease) of carbon uptake or decrease (increase) of carbon emission by land and ocean. It features negative values over most areas of land including areas of major terrestrial CO2 sinks, such as the central-eastern US, north China, and north Europe (shown in Figure 5b). It implies that the carbon uptakes by land over these areas are intensified with time. The area of large carbon source in the Amazon (Figure 5b) is depicted as an area of remarkable negative trend in Figure 13. This denotes a weakening of carbon emission with time. There are several regions of positive trends in the northwest U.S., southern Europe, and eastern Africa (Figure 13) showing a long-term trend of weakened carbon uptake by land. Increase of oceanic CO2 uptake is mainly distributed in higher latitudes of the North Atlantic, the Southern Ocean, and the North Pacific east to Japan (Figure 13). The main patterns for the trends of oceanic CO2 uptake resemble other available results, such as those of the Max Planck Institute Earth System Model [Crueger et al., 2008].

Figure 13.

Linear trend (evaluated from 1951 to 2000) of the annual mean CO2 flux at the surface (accounted positive upward) expressed as an average annual increment (g C m−2 yr−1). Negative value denotes increase (decrease) of carbon uptake (emission) rate by land or ocean.

[50] Interannual variations of air-land carbon exchange are closely related to local air temperature and soil moisture which are the most important climate variables driving the vegetation dynamics and land carbon cycle. Figure 14 shows temporal correlation coefficients of the net land-to-atmosphere carbon flux with the surface air temperature (Figure 14a) with the soil moisture (Figure 14b). A linear trend that is mainly related to the global warming is removed before calculating the correlation coefficient to focus only on the interannual timescale. Figure 14a shows clearly that the net carbon flux from land to atmosphere in low-to-midlatitudes is positively correlated with the surface air temperature. High correlation coefficients larger than +0.4 cover almost the whole continental areas between 40°S and 40°N and are significant at the 95% level. It means that the land uptake of CO2 decreases with increased air temperature. This is due to the fact that a high air temperature can limit the photosynthesis of vegetation during their growing season while increase the soil respiration at the same time. This is coherent with the positive correlation between the soil temperature and soil respiration, found in Savage and Davidson [2001] and Borken et al. [2003]. However, in high latitudes of the northern hemisphere and over the main part of the Tibetan Plateau where climatological mean surface air temperature is low and temperature is the primary constraint on vegetation growth, an increase of air temperature is more favorable for the vegetation photosynthesis and results in an increased carbon uptake by land.

Figure 14.

Temporal correlation coefficients between the annual mean net carbon flux (accounted positive from land to atmosphere) and (a) the air surface temperature, (b) the soil moisture in the 1 m layer for the period of 1951–2000 from the historical experiment. The mean trends in surface temperature, soil moisture, and carbon flux over the period of 1951–2000 are removed before calculating the correlation.

[51] The correlation between the land-to-atmosphere carbon flux and the soil moisture (Figure 14b) is almost opposite to the counterpart involving surface air temperature. Negative correlations span almost the whole land, except for the higher northern latitudes and the eastern Tibetan Plateau. The simulated land CO2 uptake intensifies with increased soil moisture in most of the globe at interannual timescale. A plausible explanation is that a wetter soil is more favorable for the increase of NPP than for the increase of soil respiration in lowland areas. The situation is converse in high altitudes like the Tibetan Plateau. Savage and Davidson [2001] have shown that a drier condition resulting from climate change likely decreases soil respiration in uplands and increases in wetlands. In other words, wetter conditions can increase soil respiration in uplands, which is consistent with the behavior of BCC-CSM1.1 over the Tibetan Plateau.

[52] Figure 15 shows relative contributions to yearly averaged net carbon flux for surface air temperature and soil moisture variations from results of linear regression. In tropical areas, soil moisture is more important than air temperature because moist soil condition is favorable for more terrestrial CO2 uptake, while high temperature impedes vegetation growth in usually hot tropical regions. But in other parts of the globe, surface air temperature is more important to land carbon dynamics.

Figure 15.

The colored areas show where the contribution of soil moisture to the variance of yearly averaged variation of net carbon flux from land larger than that of surface air temperature. They are results from the linear regression of yearly averaged net carbon flux with soil moisture and surface air temperature over the period of 1951–2000, and their mean trends are removed before calculating regression.

[53] In Figures 16-19, we present the regional mean time evolutions (from 1950 to 2000) of a few relevant variables including the surface air temperature, precipitation, soil moisture within the top 1 m layer, and the net carbon flux to the atmosphere, averaged for the Amazon tropical forest, eastern U.S., eastern China, and western Europe.

Figure 16.

Annual mean time series, averaged over the Amazon (20°S–10°N, 50°W–70°W), for (a) surface air temperature from BCC_CSM1.1 and the NCEP reanalysis (thin solid line), (b) precipitation rate from BCC_CSM1.1 and the CMAP observation (thin solid line), (c) soil moisture in the 1 m layer from BCC_CSM1.1 and the NCEP reanalysis data (thin solid line), and (d) net CO2 flux from land to atmosphere. The thick solid lines denote the simulation from the historical experiment. The units are (a) K, (b) mm d−1, (c) g g−1, and (d) g C m−2 yr−1, respectively.

Figure 17.

The same as in Figure 16 but for the regional mean in the eastern U.S. (70°W–90°W, 35°N–45°N).

Figure 18.

The same as in Figure 16 but for eastern China (100°E–120°E, 25°N–40°N).

Figure 19.

The same as in Figure 16 but for western Europe (15°E–40°E, 45°N–60°N).

[54] The Amazon tropical forest is an old-aged tropical forest containing large stores of live biomass and soil organic matter. Net carbon source to the atmosphere in the Amazon, simulated in BCC-CSM1.1, is mainly attributed to biases of the model which result in a warm and dry soil. As shown in Figure 16a, the Amazon (20°S–10°N, 50°W–70°W) regionally averaged surface air temperature in the simulation is systematically higher than in the NCEP reanalysis by about 2–3 K. From the 1960s to 2000s, there is a slight trend of warming after 1980 in BCC_CSM1.1, which does not exist in the NCEP reanalysis. There is also a dry bias of about 1.5 mm d−1 in precipitation in the Amazon in comparison to the CMAP data from 1979 to 2005 (Figure 16b). Warm temperature and less precipitation lead to dryness in the soil. As shown in Figure 16c, the Amazon regionally averaged soil moisture in BCC_CSM1.1 is about 0.11 drier than the soil moisture of 0.31 ~ 0.33 in the NCEP reanalysis. The net carbon flux to the atmosphere in the Amazon has a large interannual variation (Figure 16d), which is negatively correlated with the soil moisture at the interannual timescale (Figure 16c) with a correlation coefficient of −0.92.

[55] As for the eastern U.S. (70°W–90°W, 35°N–45°N, Figure 17a), the observed variation of regional mean surface air temperature from 1951 to 1980 is well simulated by BCC_CSM1.1, especially the cooling trend in the 1950s. After 1980, the surface air temperature is slightly overestimated by about 1 ~ 2°C by BCC_CSM1.1. In Figure 17b, the intensity of simulated precipitation is close to the CMAP data. The trend of soil moisture is also well captured in comparison with the NCEP reanalysis (Figure 17c), although the simulation is a little drier by about 0.05. Under this climate condition, the eastern U.S. is simulated as a carbon sink (Figure 17d). Our results seem coherent with Nemani et al. [2002] showing a trend of increasing terrestrial carbon uptake (due to increased precipitation and soil moisture) over North America.

[56] Eastern China (100°E–120°E, 25°N–40°N) is simulated as one of the largest carbon sink regions. As shown in Figure 18, the annual surface air temperature from the NCEP reanalysis almost falls in the range of 283 K–285 K from 1950 to 1995, and there is an evident warming trend after 1995. This tendency is well reproduced by BCC_CSM1.1, although the simulated mean surface air temperature is generally lower than that in the NCEP reanalysis by about 2°C. The regional mean annual precipitation in BCC_CSM1.1 is systematically larger than the CMAP observation by about 1 mm d−1. There is not any obvious trend from 1951 to 2005 for either BCC_CSM1.1 or CMAP. The variation of BCC_CSM1.1 simulated soil moisture is generally consistent with that in NCEP reanalysis although it is drier by about 0.02. Eastern China is a persistent carbon sink in the simulation of BCC_CSM1.1. We note however that a few recent studies showed a high vulnerability for the terrestrial ecosystem in eastern China. For example, Xiao et al. [2009] and Zhang et al. [2009] showed that severe and extended droughts can significantly affect the terrestrial carbon cycling in China and even cause the countrywide terrestrial ecosystems to switch from a carbon sink to a source.

[57] European ecosystems have been reported to be a carbon sink, which is estimated to be 135–205 GtC yr−1 [Janssens et al., 2003], 185–285 GtC yr−1 [Schulze et al., 2009], and an average of 100 GtC yr−1 between 1980 and 2007 [Churkina et al., 2010] based on the compilation of various observations. The simulated net carbon flux to the atmosphere (Figure 19d) reveals that western Europe is a persistent carbon sink in BCC_CSM1.1, consistent with the above-cited studies. The mean flux averaged for the region (15°E–40°E, 45°N–60°N) is about 70 g C m−2 yr−1. There is a slight decreasing trend in the annual mean carbon uptake in the late half of the 20th century. This trend in long-term timescale partly results from the warming in the air temperature. As shown in Figure 19a, the observed warming trend of the surface air temperature from the 1960s to 2000s averaged for western Europe is well reproduced by BCC_CSM1.1. The simulation of precipitation and soil moisture from BCC_CSM1.1 is close to the CMAP observation and NCEP reanalysis. There does not exist any evident signal of increasing or decreasing tendency in precipitation and soil moisture from 1951 to 2000.

4 Summary and Discussion

[58] The paper presented the basic performance of the Beijing Climate Center Climate System Model (BCC_CSM1.1) in reproducing the global carbon cycle from 1850 to 2005. BCC_CSM1.1 is a global ocean-atmosphere-land-ice fully coupled model with an interactive carbon cycle. A 300 year preindustrial control experiment and the historical experiment from 1850 to 2005 were conducted by BCC_CSM1.1 with prescribed anthropogenic CO2 emission and other historical forcing following the CMIP5 recommendation.

[59] BCC_CSM1.1 can well reproduce the global trend and evolution of the atmospheric CO2 concentration and surface air temperature from 1850 to 2005. There is only 5 ppmv higher than the observation of CO2 concentration at year 2005. Both land and ocean act as an important carbon sink in the 20th century. Total CO2 uptakes by the global land and ocean are 3.6 GtC yr−1 for the decades of 1980s and 4.0 GtC yr−1 for the decades of 1990s. They compared reasonably well to previous observation-based or model estimates.

[60] Regional variation of CO2 uptake in land is examined in detail. The largest terrestrial CO2 sink over the globe in BCC_CSM1.1 is distributed in the northern midlatitudes, with three significant carbon sink areas in eastern U.S., eastern China, and western Europe. The Amazon is simulated as a net carbon source to the atmosphere by BCC_CSM1.1. This is related to the fact that the Amazon is an old-aged tropical forest with large carbon storage. It seems that BCC_CSM1.1 underestimates the NPP in the Amazon, due to model biases of insufficient precipitation in this region.

[61] The carbon exchange with the atmosphere in the Amazon is still of large uncertainty. Rice et al. [2004] used the observation for a well-drained mature upland forest in the Tapajos National Forest near Santarem, Para, Brazil (2°51′S, 54°58′W) and pointed out that transfer of carbon between live and dead biomass pools can lead to substantial increases in the pool of coarse woody debris and finally cause net carbon release to the atmosphere in this region. Based on a numerical simulation using satellite observations of vegetation cover, Potter et al. [2005] predicted a CO2 source of 0.17 GtC per year in 1983 in the Amazon. In a recent study of Phillips et al. [2009], it is shown that if droughts become more frequent in some tropical regions such as in the Amazon in 2005, then the biomass sink may flip into a source. These results indicate that the Amazon may be an area of carbon source under some conditions.

[62] In the last half of the 20th century, there is an obvious increase of carbon uptake by the global land, mainly in the Amazon, eastern North America, and eastern China. However, this increasing trend of the continental carbon uptake in the 20th century cannot be used alone to draw conclusions on climate-carbon feedback that is thought to be positive in most numerical models [Friedlingstein et al., 2006], since the CO2 concentration has also largely increased during the 20th century. A precise evaluation of the climate-carbon feedback is therefore needed to analyze further appropriately designed sensitivity experiments.

[63] At interannual timescale, BCC_CSM1.1 shows a positive correlation between the net carbon flux (accounted positive from land to atmosphere) and the surface air temperature for most continental areas of low and midlatitudes. We might deduce the positive climate-carbon feedback (warmer temperature leads to smaller land carbon uptake). But we also find a negative correlation between the land carbon uptake and soil moisture, that is, a wetter soil is more favorable for a larger NPP than for a stronger soil respiration, hence a more intense terrestrial carbon uptake. We need to emphasize that such relationships are valid at interannual timescales, but maybe not for a global warming trend. The relative contribution of soil moisture to terrestrial carbon cycle over tropical regions is more important than that of air temperature, but the air temperature is more important than soil moisture over other regions of the globe.

[64] The main spatial pattern of the air-sea CO2 exchange fluxes is featured as an outgassing of CO2 to the atmosphere in the equatorial oceans, and an oceanic uptake at higher latitudes, such as the North Pacific and North Atlantic. In comparison to the natural carbon exchange in preindustrial conditions, the anthropogenic carbon uptakes by the global oceans are mainly distributed in the Southern Ocean and North Atlantic.

[65] The interannual variability of atmospheric CO2 shows remarkable correlation with ENSO. The positive correlation between the natural variation of global atmospheric CO2 and the Niño-3 index based on the observation data was reproduced in BCC-CSM1.1. This is mainly attributed to the CO2 variation in the tropics. Further studies are necessary to investigate ENSO and its impacts on the atmospheric circulation and precipitation, and consequently on the land and ocean carbon uptakes.

[66] Although several recent studies suggested that some potential physical and biogeochemical drivers of the ocean carbon cycle are favorable for a decrease of CO2 uptake in the Southern Ocean in a warmer climate [e.g., Sarmiento et al., 1998; Cox et al., 2000; Plattner et al., 2002; Russell et al., 2006; Le Quere et al., 2007], the opposite is also found by others [e.g., Zickfeld et al., 2007; McNeil et al., 2001; Crueger et al., 2008; Matear and Lenton, 2008], which supports our findings of an increase of oceanic carbon uptake because of the anthropogenic CO2 emission during the 20th century. For example, McNeil et al. [2001] showed an increase in observed CO2 uptake for the sub-Antarctic oceanic zone between 45°S and 50°S during a time interval of 28 years. Their estimation of the CO2 uptake ranged from 0.73 to 0.86 µmol kg yr−1 between 1968 and 1996.

[67] The carbon flux in the Southern Ocean is a complex issue. Matear and Lenton [2008], by using an ocean biogeochemical model, concluded that the CO2 uptake in the Southern Ocean remains almost unchanged when climate changes, since two opposite effects almost cancel out each other: an increase in heat and freshwater fluxes can lead to a net increase in the Southern Ocean uptake (south of 40°S), while an increase in wind stresses lead to a net decrease in uptake. As pointed out by Ito et al. [2010], the mechanism and pathways of anthropogenic CO2 uptake and transport are poorly understood. Causes of the increase of CO2 uptake in the Southern Ocean in BCC_CSM1.1 deserve further investigation.

[68] In this work, anthropogenic land cover change (LCC) is not explicitly included in the land model of BCC_CSM1.1. But the prescribed total anthropogenic CO2 emission, following the CMIP5 recommendation, does take into account the LCC in the historical simulation. As a result, the radiative effect of CO2 emission from LCC on the climate is, at least in part, implicitly taken into account in our simulation, albeit the following two processes are missing in the model: (a) the decrease in the sink capacity of the global terrestrial biosphere due to reduction of the residence time of carbon when, for example, forests or grasslands are converted to cultivated land [e.g., Gitz and Ciais, 2004]; and (b) the effect of anthropogenic land cover change on climate through changes in the physical properties of the land surface [e.g., Brovkin et al., 2004, 2006; Betts et al., 2007]. The latter process, also neglected in many previously published climate-carbon cycle models, is now included in some (not all) CMIP5 models and will surely be among our objectives in future model development.

[69] We also noticed relatively high carbon sink/source in regional scales after 300 year control run. It is possibly caused by relatively short simulation period to reach the model equilibrium, or existence of the multidecadal to centennial scale drifts at regions in BCC_CSM1.1. More experiments are needed to understand the centennial scale drift in fully coupled climate models in the future.

Acknowledgments

[70] This work was supported by National Basic Research Program of China (973 Program:2010CB951902) and Special Program for China Meteorology Trade (Grant No. GYHY201306020).

Ancillary