-
The model data used in this study is the simulation results of historical experiments of 12 ESMs from CMIP6 and 11 ESMs from CMIP5. Table 1 shows model information for those in CMIP6 and Table 2 for those in CMIP5. The analyzed variables include the near-surface CO2 concentration and anthropogenic and natural CO2 fluxes. Anthropogenic CO2 fluxes comprise those due to anthropogenic emissions, including the combustion of fossil fuels, cement production, agricultural practices, and sources associated with anthropogenic land-use changes (excluding forest regrowth). Natural CO2 fluxes are the sum of land–atmosphere CO2 exchanges (i.e., net ecosystem productivity) and ocean–atmosphere CO2 exchanges. The analysis and discussion of anthropogenic and natural CO2 fluxes are based on output from BCC-CSM2-MR, MRI-ESM2-0, and UKESM1-0-LL since these models provided all the necessary flux variables. Due to the different resolutions used in these ESMs, all model output is interpolated to a global grid of 1.125° × 1.125° with bilinear interpolation to facilitate analysis and comparison. The CMIP6 model data are interpolated from the model grids to each site using the site observations.
ESMs Institute Number of ensemble members Resolution
(lon × lat)Carbon/Biogeochemistry Reference Land Ocean ACCESS-ESM1-5 Commonwealth Scientific and Industrial Research Organization, Australia 10 1.875° × 1.25° CABLE2.4 with CASA-CNP WOMBAT Ziehn et al., 2020 BCC-CSM2-MR Beijing Climate Center, China 3 1.125° × 1.125° BCC-AVIM2 MOM4_L40 Wu et al., 2019 CanESM5 Canadian Centre for Climate Modelling and Analysis, Canada 15 2.8125° × 2.8125° CLASS-CTEM CMOC (biol) Swart et al., 2019 CESM2 National Center for Atmospheric Research, United States 4 1.25° × 0.9375° CLM5 MARBL Danabasoglu et al., 2020 CNRM-ESM2-1 Centre National de Recherches Meteorologiques, France 4 1.406° × 1.406° ISBA-CTRIP PISCESv2-gas Séférian et al., 2019 EC-Earth3-CC EC-Earth-Consortium, Europe 1 3° × 2° HTESSEL and LPJ-GUESS v4 PISCES v2 Döscher et al., 2022 GFDL-ESM4 National Oceanic and Atmospheric Administration/Geophysical Fluid Dynamics Laboratory, United States 1 1.25° × 1° LM4.1 COBALTv2 Dunne et al., 2020 MIROC-ES2L Japan Agency for Marine-Earth Science and Technology, Japan 3 2.8125° × 2.8125° MATSIRO (phys) and VISIT-e (BGC) OECO2 Hajima et al., 2020 MPI-ESM1-2-LR Max Planck Institute for Meteorology, Germany 10 1.875° × 1.875° JSBACH3.2 HAMOCC6 Mauritsen et al., 2019 MRI-ESM2-0 Meteorological Research Institute, Japan 1 1.125° × 1.125° HAL 1.0 MRI.COM4.4 Yukimoto et al., 2019 NorESM2-LM Norwegian Climate Center, Norway 2 2.5° × 1.875° CLM5 HAMOCC5.1 Seland et al., 2020 UKESM1-0-LL Met Office Hadley Centre, UK 4 1.875° × 1.25° JULES-ES-1.0 MEDUSA-2.1 Sellar et al., 2019 Table 1. Information on the 12 ESMs participating in CMIP6 in this study.
ESMs Institute Number of ensemble members Resolution
(lon × lat)Carbon/Biogeochemistry Reference Land Ocean BCC-CSM1-1 Beijing Climate Center, China 1 2.8125° × 2.8125° BCC-AVIM1.0 MOM4_L40 Wu et al., 2013 BNU-ESM Beijing Normal University, China 1 2.8125° × 2.8125° CoLM3 and BNU-DGVM (C/N) MOM4p1 and iBGC Ji et al., 2014 CanESM2 Canadian Centre for Climate Modelling and Analysis, Canada 3 2.8125° × 2.8125° CTEM CMOC Arora et al., 2011 CESM1-BGC National Center for Atmospheric Research, United States 1 1.25° × 0.9375° CLM4 POP2 and BEC Long et al., 2013 FIO-ESM The First Institute of
Oceanography, China1 2.8125° × 2.8125° CLM3.5 and CASA POP2.0 and OCMIP-2 Qiao et al., 2013 GFDL-ESM2G Geophysical Fluid Dynamics Laboratory, United States 1 2.5° × 2° LM3.0 GOLD and TOPAZ2 Dunne et al., 2012, 2013 GFDL-ESM2M Geophysical Fluid Dynamics Laboratory, United States 1 2.5° × 2° LM3.0 MOM4p1 and TOPAZ2 Dunne et al., 2012, 2013 MIROC-ESM Japan Agency for Marine-Earth Science and Technology, Japan 3 2.8125° × 2.8125° MATSIRO and SEIB-DGVM COCO3.4 and NPZD Watanabe et al., 2011 MPI-ESM-LR Max Planck Institute for Meteorology, Germany 3 1.875° × 1.875° JSBACH MPIOM and HAMOCC5 Giorgetta et al., 2013 MRI-ESM1 Meteorological Research Institute, Japan 1 1.125° × 1.125° HAL MRI.COM3 Yukimoto et al., 2012; Adachi et al., 2013 NorESM1-ME Norwegian Climate Center, Norway 1 2.5° × 1.875° CLM4 MICOM and HAMOCC5 Tjiputra et al., 2013 Table 2. Information on the 11 ESMs participating in CMIP5 in this study.
The reference data used to evaluate the performance of the models’ historical simulations includes a) the global site observation data of monthly mean atmospheric CO2 concentration from 1968 to 2021 provided by the World Data Centre for Greenhouse Gases (WDCGG) (WMO, 2021) at the 49 representative sites (shown by the triangles in Fig. 1) which have continuous observations from 1995 to 2014, b) the monthly mean data of zonal mean observed atmospheric CO2 concentration (Meinshausen et al., 2017) and global anthropogenic CO2 emissions (Hoesly et al., 2018) from 1850 to 2014, as recommended by CMIP6, c) the annual observed atmospheric CO2 concentration data from 1850 to 2014 provided by the Scripps CO2 program, which are obtained by combining ice core data from 1850 to 1958 (MacFarling Meure et al., 2006) and data observed in Mauna Loa and the South Pole from 1959 to 2014 (Keeling et al., 2001), and d) the global carbon budget data from 1850 to 2014 (GCB2021; Friedlingstein et al., 2022).
Figure 1. Geographical location of 154 WDCGG observation sites (WMO, 2021). The 49 sites marked with red triangles are representative sites selected for this study with continuous monthly CO2 concentration data for 1995–2014.
-
Figure 2a shows the time series of the global annual mean surface CO2 concentration from 1850 to 2014, as simulated by CMIP6 ESMs. The reference data is the time series of the Scripps CO2 concentration and the CMIP6 recommended CO2 concentration based on observational and reconstructed data. The two sets of reference data have a high level of agreement on the concentration and evolution of CO2, as the correlation coefficient exceeds 0.99. The average difference between the multi-model ensemble mean (MME) and the reference data is within 2.5 ppmv, and the MME can reasonably simulate both the slow increases in observed CO2 concentration before the 1950s and the more rapid growth thereafter. The 12 ESMs can also basically reproduce the trend of CO2 concentration change, but the differences in CO2 concentration between the ESMs show an increase with time. The simulated CO2 concentrations range from 282.2 to 291.4 ppmv (an ensemble spread of 9.2 ppmv), in 1850, to between 368.5 and 432.1 ppmv (a spread of 63.6 ppmv), in 2014. Table 3 shows the changes and trends in CO2 concentration simulated by each ESM during the period of rapidly increasing CO2 (1950–2014). Although the MME long-term trend is in good agreement with the observation, the inter-model differences in CO2 concentrations increase with time from 27.7 ppmv in 1950 to 63.6 ppmv in 2014. This leads to a large range of simulated trends (1.00 to 1.57 ppmv yr–1). In contrast, the CMIP5 ESMs systematically overestimate the CO2 concentration after 1950 (Fig. 2b), which is consistent with the results of Hoffman et al. (2014) and Friedlingstein et al. (2014). These biases are much smaller in CMIP6 ESMs, which may be due to the more comprehensive biogeochemical processes included in CMIP6 ESMs, such as nitrogen and nutrient limitations on land plant growth and iron limitations in ocean ecosystems (Canadell et al., 2021).
Figure 2. Time series of the globally averaged annual mean of CO2 concentration during 1850–2014 for (a) CMIP6 and 1850–2005 for (b) CMIP5: ESMs (colored thin solid lines), MME (red thick solid line), the observation data provided by Scripps CO2 program (gold thick solid line) and the CMIP6 recommended observation data (black thick solid line). Units: ppmv.
Data sources 1950
(ppmv)2014
(ppmv)Trend
(ppmv yr–1)ACCESS-ESM1-5 308.4 395.2 1.31 BCC-CSM2-MR 299.6 368.5 1.03 CanESM5 307.7 409.2 1.56 CESM2 321.3 416.0 1.43 CNRM-ESM2-1 305.0 374.4 1.00 EC-Earth3-CC 327.3 432.1 1.54 GFDL-ESM4 315.8 419.7 1.57 MIROC-ES2L 314.4 386.5 1.05 MPI-ESM1-2-LR 311.7 406.2 1.45 MRI-ESM2-0 307.3 381.3 1.16 NorESM2-LM 315.0 406.4 1.39 UKESM1-0-LL 315.0 413.2 1.52 MME 312.4 400.7 1.33 Observation (CMIP6) 312.8 397.5 1.36 Table 3. The globally averaged annual mean CO2 concentrations (ppmv) in 1950 and 2014 and the trends for 1950–2014 (ppmv yr–1) simulated by 12 CMIP6 ESMs and MME compared with the CMIP6 recommended observations.
Figure 3 shows the time evolution of the zonally-averaged annual mean of CO2 concentration anomalies for the CMIP6 reference data and ESMs simulation from 1850 to 2014. The average CO2 concentrations along each latitude circle for both CMIP6 reference data and ESMs simulations increase with time, consistent with the global average time series. The growth rate of CO2 concentration in the northern hemisphere (NH) is slightly higher than that in the southern hemisphere (SH). The change of zonal mean CO2 concentration simulated by the MME is consistent with observations (Fig. 3b), but the ensemble spread (e.g., twice the standard deviation, 2σ) increases from 3 ppmv in 1850 to over 40 ppmv in 2014 (Fig. 3c). For all latitudes, the zonal mean CO2 concentration shows remarkable increases from the year 1950. Hence, we estimate the trends of the zonally averaged annual mean of CO2 concentration from 1950 to 2014 relative to 1850–1879. As shown in Fig. 4, the highest CO2 concentration growth rate values in the NH are between 30°–50°N. As for the CMIP6 reference data, the average growth rate of CO2 concentration in the NH from 1950 to 2014 is estimated to be about 1.39 ppmv yr–1, and its SH counterpart is close to 1.33 ppmv yr–1. The CO2 concentration growth rate for the MME is slightly lower than the CMIP6 reference data by 0.03 ppmv yr–1. The MME can reasonably reproduce the main features of the meridional distribution of CO2 concentration trends with high (low) rates of increase in the northern (southern) hemisphere, with the largest values located in the mid-latitudes in the NH. The ESMs can also reasonably simulate the meridional distribution of CO2 concentration trends, indicating that the model biases are mainly due to a systematic deviation in CO2 concentration trends on the global scale.
Figure 3. Time evolution of zonally averaged annual mean of CO2 concentration for 1850–2014 relative to 1850–1879: (a) the CMIP6 recommended observation data, (b) MME, and (d)–(o) ESMs. Panel (c) shows the model spread among the 12 ESMs expressed as twice the standard deviation (2σ). Units: ppmv.
Figure 4. Trends of the zonally averaged annual mean of CO2 concentration for 1950–2014 relative to 1850–1879: ESMs (colored thin solid lines), MME (red thick solid line), and the CMIP6 recommended observation data (black thick solid line). Units: ppmv yr–1.
The inter-model differences in the historical evolution of CO2 concentration in ESMs (as shown in Figs. 2 and 3) are mainly attributed to their biases in simulated CO2 fluxes. Variation of the global mean of CO2 concentration in the atmosphere is generally determined by anthropogenic CO2 emissions and natural CO2 exchanges between land and atmosphere and between ocean and atmosphere, which are generally presented in terms of CO2 flux. Figure 5 shows the yearly global mean CO2 concentration and CO2 flux to the atmosphere for BCC-CSM2-MR, MRI-ESM2-0, and UKESM1-0-LL from 1850 to 2014, in which negative values of CO2 flux denote uptake by terrestrial or marine ecosystems. Only these three models provided CO2 flux data to download from the CMIP6 website (https://esgf-node.llnl.gov/search/cmip6/). Among the three models, the values of net CO2 flux (i.e., the sum of anthropogenic CO2 emissions and natural CO2 exchange with land or ocean) in BCC-CSM2-MR are relatively low, corresponding to low atmospheric CO2 concentration; the net CO2 flux in UKESM1-0-LL is relatively high after the 1930s, corresponding to the higher atmospheric CO2 concentration in the second half of the 20th century (Figs. 5a, b).
Figure 5. Time series of the globally averaged annual mean of (a) CO2 concentration, (b) net CO2 flux (9-yr running average), (c) net CO2 flux, (d) anthropogenic CO2 flux, (e) land-atmosphere CO2 flux, and (f) ocean-atmosphere CO2 flux for 1850–2014: BCC-CSM2-MR (deep-pink thin solid lines), MRI-ESM2-0 (cyan thin solid lines), UKESM1-0-LL (purple thin solid lines), the CO2 observation data provided by Scripps CO2 program (gold thick solid line), the CMIP6 recommended CO2 observation data (black thick solid line), the CO2 flux estimated by GCB2021 (GATM is the growth rate of atmospheric CO2 concentration) (dark blue thick solid lines), and the CMIP6 recommended anthropogenic CO2 emissions (gray thick solid line). Positive values in panels (b)–(f) represent a flux to the atmosphere. Units: ppmv for CO2 concentration and GtC yr–1 for CO2 flux.
As shown in Fig. 5, both the globally averaged CO2 concentration and CO2 flux show rapid increases after 1960. The main source of increased CO2 in the atmosphere is anthropogenic CO2 emissions. The anthropogenic emissions of CO2 into the atmosphere are over 4 GtC yr–1 (Fig. 5d), although their values vary slightly between the ESMs. These differences likely arise from the pre-processing of CMIP6-prescribed emissions to the resolution of the models. The anthropogenic CO2 emissions in Fig. 5d exceed the total CO2 uptake by the land and ocean and cause a net CO2 flux into the atmosphere almost every year (Fig. 5c). Referring to the calculation method of Friedlingstein et al. (2022), we quantified the 1960–2014 averaged partitioning of anthropogenic CO2 emissions burden into the atmosphere (AF), the land (LF), and the ocean (OF) from the three ESMs (Table 4). Compared with GCB2021 (Friedlingstein et al., 2022), the AF value simulated by BCC-CSM2-MR and UKESM1-0-LL is about 3% lower, while the LF value is about 5% higher; the LF value simulated by MRI-ESM2-0 is 6% lower, while the OF value is 7% higher. BCC-CSM2-MR and UKESM1-0-LL reproduce a stronger CO2 uptake by terrestrial ecosystems than by marine ecosystems, as derived by GCB2021.
Data sources AF LF OF ESMs BCC-CSM2-MR 41% 35% 24% MRI-ESM2-0 45% 23% 32% UKESM1-0-LL 40% 34% 25% GCB2021 44% 29% 25% Table 4. The partitioning of anthropogenic CO2 emissions in the atmosphere (airborne fraction, AF), land (land-borne fraction, LF), and ocean (ocean-borne fraction, OF) for 1960–2014 from 3 ESMs and GCB2021 (Friedlingstein et al., 2022).
It is noted that land-atmosphere CO2 fluxes exhibit much more remarkable year-to-year variations than the ocean–atmosphere CO2 fluxes. The average amplitude of variations in land–atmosphere CO2 flux among ESMs is 1.8 GtC yr–1, which is about three times that in ocean–atmosphere CO2 flux (Figs. 5e, f). Given the long lifetime of CO2, the strong interannual variability of land–atmosphere CO2 flux may lead to a cumulative increase in the inter-model differences of CO2 concentration. Therefore, the uncertainty of the net CO2 flux mainly comes from the land–atmosphere CO2 flux, which is consistent with the conclusions of IPCC AR5 and AR6 (Ciais et al., 2013; Friedlingstein et al., 2014; Canadell et al., 2021).
-
Exploring the global spatial distribution of CO2 concentration is helpful for understanding the uncertainties in ESMs. Utilizing the relatively abundant observational data from recent years, we compared the site-based observations and ESM simulations of the annual-mean atmospheric CO2 concentrations from 1995–2014 (Fig. 6). From the site observations, Europe, East Asia, and the eastern United States can be considered as regions with high CO2 concentration (e.g., 5–10 ppmv above the global average) (Fig. 6a). The MME reproduces the distribution of high-CO2 concentration in these regions (Fig. 6b). All ESMs can simulate the meridional and regional features of CO2 concentration (Figs. 6d–o), but the spatial variations of CO2 concentration are quite different among the ESMs. The model spread (2σ) of CO2 concentration differences is higher than 10 ppmv in northern Russia, Eastern China, the eastern United States, northern South America, and southern Africa (Fig. 6c). It seems that these inter-model differences are mainly due to the contributions of two model outliers: MIROC-ES2L and UKESM1-0-LL.
Figure 6. Global distribution of annual mean CO2 concentration anomalies relative to the global mean for 1995–2014: (a) site observations, (b) MME, and (d)–(o) ESMs. Panel (c) shows the model spread among the 12 ESMs expressed as twice the standard deviation (2σ). The global mean is marked at the top-right corner of each panel. Units: ppmv.
Figure 7 shows the multi-model ensemble mean simulation and the model spread of the spatial distribution of CO2 concentration differences for 1995–2005 in CMIP6 and CMIP5 ESMs. BCC-CSM1-1 is not included in the CMIP5 ensemble as it does not consider the spatial distribution of atmospheric CO2, and only the global mean of the carbon budget is considered (Wu et al., 2013). Both the CMIP6 MME and the CMIP5 MME can simulate the north-south difference of CO2 concentration and the three high-CO2 concentration centers over Europe, eastern China, and the eastern United States (Figs. 7a, b). Compared with CMIP6, although the global mean CO2 concentration is about 8 ppmv larger in the CMIP5 MME, the spatial distribution of CO2 concentration is relatively smoother with smaller CO2 concentration over the three high CO2 concentration centers. The model uncertainties of CO2 concentration anomalies are large in the three high-CO2 concentration centers in the NH in CMIP6 and CMIP5 ESMs. In addition, the model uncertainties are also large over the tropical rain forests, such as tropical Africa, Southeast Asia, and South America (Figs. 7c, d).
Figure 7. Global distribution of annual mean CO2 concentration anomalies relative to the global mean for 1995–2005: (a) CMIP6 MME, (b) CMIP5 MME (BCC-CSM1-1 is not included). The model spread among (c) the 12 CMIP6 ESMs and (d) the 10 CMIP5 ESMs expressed as twice the standard deviation (2σ). The global mean is marked at the top-right corner of each panel. Units: ppmv.
The model uncertainties of CO2 concentration over the ocean in ESMs, which may be mainly caused by the difference in ocean–atmosphere CO2 fluxes among ESMs, are less than those over land. As described in Wu et al. (2013), the air-sea CO2 flux depends on the difference between the CO2 concentration of the water vapor-saturated air and the sea-surface aqueous CO2 concentration, and the magnitude of wind speed near the surface. The simulation deviation of sea surface CO2 concentration due to the biases of sea surface temperature (SST), dissolved inorganic carbon, and total alkalinity may be the main reason for the difference in the simulations of air-sea CO2 flux (Jing et al., 2022).
Noticeable simulation uncertainty of CO2 concentration over land may be mainly related to the large difference in land–atmosphere CO2 fluxes among ESMs. The land–atmosphere CO2 flux mainly involves the net change in carbon storage in the ecosystem, commonly called net ecosystem productivity (NEP), including fluxes from vegetation, detritus, and mineral soil. It is just the residual of gross primary productivity (GPP) of terrestrial vegetation by deducting the plant autotrophic respiration (Rveg) and the heterotrophic soil respiration (Rsoil). The magnitude of NEP is less than that of all the three terms, GPP, Rveg, and Rsoil, and easily introduces a large uncertainty to the land–atmosphere CO2 flux simulations among ESMs, within which different treatments are used in land CO2 schemes. As listed in Table 5.4 of IPCC AR6 on the characteristics of land carbon cycle components of most CMIP6 ESMs (Arora et al., 2020; Canadell et al., 2021), the number of carbon pools and the number of plant functional types in the land CO2 schemes are different among the ESMs. Only a few ESMs represent fire, dynamic vegetation cover, permafrost carbon, and nitrogen cycles in land CO2 schemes, and most ESMs do not or just partially represent them. The simulated deviation of climate (such as temperature, precipitation, etc.) in ESMs may also cause large uncertainties in the simulation of terrestrial carbon flux by affecting these terrestrial carbon cycle processes (Ahlström et al., 2017; Bonan et al., 2019).
A comparison of the simulated CO2 concentration with the data at 49 observation sites in the globe (Fig. 8) shows that the mean difference between the MME results and the observations (2 ppmv) is smaller than that between each ESM and the observations (–27.3 to 28.8 ppmv). As shown in Figs. 8b–m, large model biases exist in MIROC-ES2L and UKESM1-0-LL, and their spatial standard deviations reach 6.8 and 7.2 ppmv, respectively, while the other ESMs range around 2–3 ppmv. Meanwhile, the correlation coefficients (0.86–0.88) between these two models and the site observations are also relatively lower than those between other models and site observations (0.98–0.99).
Figure 8. Comparison of the annual mean CO2 concentrations from 1995 to 2014 at 49 observation sites with those simulated by (a) MME and (b–m) ESMs at the corresponding sites. Here, the model simulation at each site shows the data interpolated from the model grids to each site. The black dashed line shows a 1:1 straight line for reference. Each panel also shows the mean and uncertainty (±1 standard deviation, ±1σ) of the deviations (DEV) between simulated and observed CO2 concentrations and the correlation coefficient (r) between simulation and observation. Units: ppmv.
-
At regional scales, the CO2 concentration can have strong seasonal cycles; thus, we calculated the seasonal amplitude of atmospheric CO2 concentration, defined as the difference between the maximum and minimum monthly mean CO2 concentration during a year. Figure 9 shows the temporal evolution of the seasonal amplitude of zonal mean CO2 concentration for 1850–2014 from CMIP6 reference data and ESMs simulations. Similar to the zonal distribution of CO2 concentration, the seasonal amplitude of CO2 concentration in the NH is generally higher than that in the SH and has an increasing trend after the 1950s, with the largest trends located near 60°N (Fig. 9a). Compared with the CMIP6 reference data, the MME reasonably reproduces the meridional variations of the seasonal amplitude of zonal mean CO2 concentration and the increasing amplitudes in the later period. The largest seasonal variation of CO2 occurs near 60°N, with increases of more than 6 ppmv by the 2010s compared with the 19th century (Fig. 9b). Most ESMs can also capture the meridional difference and increasing seasonal amplitudes of zonal mean CO2 concentration (Figs. 9d–o). However, there is a remarkable spread across the ESMs. As shown in Fig. 9c, the spread (2σ) exceeds 24 ppmv in the vicinity of 40°–70°N, significantly higher than the MME values. This means that the ESMs simulation of the seasonal amplitude of CO2 concentration is highly uncertain. As the mean of three ESMs (BCC-CSM2-MR, MRI-ESM2-0, and UKESM1-0-LL) in Fig. 10 shows, the seasonal amplitude increase of natural CO2 fluxes is much larger than that of anthropogenic CO2 fluxes after 1950, especially around 60°N, so the increase in seasonal amplitude of atmospheric CO2 concentration is mainly related to seasonal variations of natural CO2 fluxes. In Fig. 9, the increase in seasonal amplitude of CO2 concentration with time may be strongly correlated with the amplified vegetation productivity in the NH caused by the interaction of recent climate change and vegetation dynamics that have been suggested by previous studies (Graven et al., 2013; Forkel et al., 2016; Piao et al., 2018; Bastos et al., 2019).
Figure 9. Time evolution of the seasonal amplitude of zonal mean CO2 concentrations for 1850–2014: (a) CMIP6 recommended observation data, (b) MME, and (d)–(o) ESMs. Panel (c) shows the model spread among the 12 ESMs expressed as twice the standard deviation (2σ). Units: ppmv.
Figure 10. The increase in the seasonal amplitude of zonal mean (a) anthropogenic CO2 flux, (b) natural CO2 flux, and (c) net CO2 flux of ensemble mean of BCC-CSM2-MR, MRI-ESM2-0, and UKESM1-0-LL for 1950–2014 relative to 1950. Units: gC m–2 mon–1.
We further explored the global distribution of the annual mean of the seasonal amplitude of CO2 concentrations for 1995–2014 (Fig. 11). Observations and MME results show that the seasonal amplitude of CO2 concentration over land is higher than that over the ocean and that the mid-high latitudes of the NH, such as Europe, Eastern China, eastern United States, and Alaska, are areas with high seasonal amplitude of CO2 concentration (Figs. 11a, b). MME results show that the seasonal amplitudes are higher than 22 ppmv in Europe, North Asia, eastern China, and eastern North America and higher than 12 ppmv in northern South America and southern Africa. Most ESMs can simulate the meridional and regional differences in the seasonal amplitude of CO2 concentrations (Figs. 11d–o). However, there are some notable exceptions: for example, compared with site observations, MIROC-ES2L overestimates the seasonal amplitude of CO2 concentration by 30 ppmv in the NH, Southern Africa, and South America.
Figure 11. Global distribution of the annual mean of the seasonal amplitude of CO2 concentrations for 1995–2014: (a) site observations, (b) MME, and (d)–(o) ESMs. Panel (c) shows the model spread among the 12 ESMs expressed as twice the standard deviation (2σ). The global mean is marked at the top-right corner of each panel. Units: ppmv.
The regions with high seasonal-amplitude values of CO2 concentration correspond to those areas that show a large model spread (Fig. 11c), further noting that the value of the model spread is even larger than that of the seasonal amplitude of CO2 concentration in the MME results (Fig. 11b). In general, the seasonal amplitude of CO2 concentration and its model spread over oceans are smaller than those over land at the same latitude; the seasonal amplitude of CO2 concentration and its model spread over land in the NH are higher than those in the SH.
We further examine the mean seasonal variation of CO2 concentration for 1995–2014 (Fig. 12) in the NH, SH, and eight major sub-regions (Fig. 11a). MIROC-ES2L was not included in the analysis because the simulated amplitude is anomalously strong compared with observations. The observed seasonal amplitude of CO2 concentration in the NH is 11.7 ppmv (Fig. 12a) and about 4.1 ppmv in the SH (Fig. 12b). The relatively large portion of the SH covered by oceans may be one of the important factors for the relatively small seasonal variability of CO2 concentration there. In the NH, winter and summer are the peak and trough periods of CO2 concentration, respectively. This seasonal cycle is mainly caused by the release of CO2 from the decomposition of soil organic matter in winter and the absorption of CO2 by the photosynthesis of terrestrial ecosystems in summer (Keeling et al., 1996; Wenzel et al., 2016). The seasonal cycles of CO2 concentration simulated by the MME are generally consistent with the site observations, but the amplitude is slightly smaller, and the period of low values in the year is slightly earlier than the observations. Both observations and the MME show that the seasonal amplitudes of CO2 concentrations in Europe, North Asia, and northern North America are generally higher than those in East Asia and southern North America (Figs. 12c–g). Notably, the MME results over the SH show that the seasonal amplitudes in South America and southern Africa are about 5 ppmv, and those in Australia are only around 2.5 ppmv (Figs. 12h–j).
Figure 12. Mean seasonal cycle (expressed as monthly anomalies) of CO2 concentration for 1995–2014: (a) Northern Hemisphere, (b) Southern Hemisphere, (c) Europe (35°–70°N, 0°–50°E), (d) North Asia (50°–70°N, 60°–150°E), (e) East Asia (25°–50°N, 90°–150°E), (f) Northern North America (50°–72°N, 165°–60°W), (g) Southern North America (25°–50°N, 120°–60°W), (h) South America (55°S–0°, 75°–45°W), (i) Australia (45°–10°S, 115°–155°E), and (j) Southern Africa (40°S–0°, 10°–40°E). Colored thin solid lines, red thick solid lines, and black thick solid lines represent the ESMs, MME, and observations, respectively. The observation values are the average of all sites in the selected area. MIROC-ES2L was excluded due to the large deviation of the simulated seasonal amplitude from the observation. Each panel also shows the mean and uncertainty (±1 standard deviation, ±1σ) of the simulated seasonal amplitude and the observed seasonal amplitude. Units: ppmv.
It is worth noting that the spatial distribution of the simulated seasonal amplitude of CO2 concentration (Fig. 11b) is not only consistent with that of annual-mean CO2 concentration (Fig. 6b) but is also correlated with the global distribution of forests (Hansen et al., 2013) and the type of vegetation (Olson et al., 2001). In ESMs, the seasonal amplitudes over the mid- and high-latitude forests in the NH are generally higher than those in the broad-leaved forest belts in East Asia and the eastern United States (Figs. 11d–o; Figs. 12c–g); the desert areas in North Africa, West Asia, and the western United States, and the Qinghai-Tibet Plateau are areas with low seasonal amplitudes (Figs. 11d–o); as the land in the SH is mainly covered by tropical rain forests, tropical grasslands, and deserts, the seasonal amplitudes of CO2 concentrations in these areas are also smaller (Figs. 11d–o; Figs. 12h–j).
-
The simulation of atmospheric CO2 concentration is closely related to the CO2 fluxes. Figure 13 shows the MME results of the annual mean and seasonal amplitude of anthropogenic CO2 flux, natural CO2 flux, and net CO2 flux for 1995–2014 (colored figure), as well as the zonal mean for three ESMs (BCC-CSM2-MR, MRI-ESM2-0, and UKESM1-0-LL) (colored curves). The high-value areas of anthropogenic CO2 emissions are located around 45°N, and the highest values are mainly in Europe, eastern China, and the eastern United States (Fig. 13a). The MME results show that land areas, other than deserts (such as the Sahara Desert) and high-altitude areas (such as the Qinghai-Tibet Plateau and the Rocky Mountains), are natural CO2 sinks, and most ocean areas except for tropical oceans are also CO2 sinks (Fig. 13b), which is consistent with the results of Takahashi et al. (2009) and Friedlingstein et al. (2022). From the meridional distribution of the zonal mean, the simulated difference of natural CO2 flux is significantly larger than that of anthropogenic CO2 flux, which is the main factor causing the inter-model differences in the net CO2 flux (Figs. 13a–c).
Figure 13. The global distribution of the annual means of (a) anthropogenic CO2 flux, (b) natural CO2 flux, and (c) net CO2 flux of the ensemble mean of BCC-CSM2-MR, MRI-ESM2-0 and UKESM1-0-LL for 1995–2014 are shown. The global distribution of the annual means of the seasonal amplitude of (d) anthropogenic CO2 flux, (e) natural CO2 flux, and (f) net CO2 flux of the ensemble mean of BCC-CSM2-MR, MRI-ESM2-0 and UKESM1-0-LL for 1995–2014 are shown. The right parts of panels (a)–(c) and panels (d)–(f) represent the meridional distribution of the zonal mean of CO2 fluxes and their seasonal amplitudes, respectively: BCC-CSM2-MR (deep-pink thin solid lines), MRI-ESM2-0 (cyan thin solid lines), UKESM1-0-LL (purple thin solid lines) and the MME (black thick solid lines). The global mean value is marked at the top-right corner of each panel. Positive values represent a flux to the atmosphere for CO2 flux. Units: gC m–2 yr–1 for CO2 flux and gC m–2 month–1 for the seasonal amplitude of CO2 flux.
As shown in Fig. 13c, yearly mean net CO2 fluxes in the NH mid-latitudes over land except for the Tibetan Plateau and the Mongolia Plateau, and in the subtropical regions of Southern Africa and South America, are mainly due to anthropogenic emissions, while in other regions they mainly come from natural contributions. But the seasonal variation of CO2 concentration is substantially determined by the seasonal cycles of natural CO2 fluxes (Figs. 13d–f), and the highest values of seasonal amplitude of CO2 concentration well correspond with those of the seasonal amplitudes of net CO2 fluxes (Fig. 11, Fig. 13f). There are weak seasonal variations in anthropogenic CO2 emissions data that are provided for the CMIP6 experiments. Therefore, the simulation difference in the seasonality of natural CO2 fluxes is the main reason for the simulation uncertainty in the seasonal variation of CO2 concentration among CMIP6 ESMs.
ESMs | Institute | Number of ensemble members | Resolution (lon × lat) | Carbon/Biogeochemistry | Reference | |
Land | Ocean | |||||
ACCESS-ESM1-5 | Commonwealth Scientific and Industrial Research Organization, Australia | 10 | 1.875° × 1.25° | CABLE2.4 with CASA-CNP | WOMBAT | Ziehn et al., 2020 |
BCC-CSM2-MR | Beijing Climate Center, China | 3 | 1.125° × 1.125° | BCC-AVIM2 | MOM4_L40 | Wu et al., 2019 |
CanESM5 | Canadian Centre for Climate Modelling and Analysis, Canada | 15 | 2.8125° × 2.8125° | CLASS-CTEM | CMOC (biol) | Swart et al., 2019 |
CESM2 | National Center for Atmospheric Research, United States | 4 | 1.25° × 0.9375° | CLM5 | MARBL | Danabasoglu et al., 2020 |
CNRM-ESM2-1 | Centre National de Recherches Meteorologiques, France | 4 | 1.406° × 1.406° | ISBA-CTRIP | PISCESv2-gas | Séférian et al., 2019 |
EC-Earth3-CC | EC-Earth-Consortium, Europe | 1 | 3° × 2° | HTESSEL and LPJ-GUESS v4 | PISCES v2 | Döscher et al., 2022 |
GFDL-ESM4 | National Oceanic and Atmospheric Administration/Geophysical Fluid Dynamics Laboratory, United States | 1 | 1.25° × 1° | LM4.1 | COBALTv2 | Dunne et al., 2020 |
MIROC-ES2L | Japan Agency for Marine-Earth Science and Technology, Japan | 3 | 2.8125° × 2.8125° | MATSIRO (phys) and VISIT-e (BGC) | OECO2 | Hajima et al., 2020 |
MPI-ESM1-2-LR | Max Planck Institute for Meteorology, Germany | 10 | 1.875° × 1.875° | JSBACH3.2 | HAMOCC6 | Mauritsen et al., 2019 |
MRI-ESM2-0 | Meteorological Research Institute, Japan | 1 | 1.125° × 1.125° | HAL 1.0 | MRI.COM4.4 | Yukimoto et al., 2019 |
NorESM2-LM | Norwegian Climate Center, Norway | 2 | 2.5° × 1.875° | CLM5 | HAMOCC5.1 | Seland et al., 2020 |
UKESM1-0-LL | Met Office Hadley Centre, UK | 4 | 1.875° × 1.25° | JULES-ES-1.0 | MEDUSA-2.1 | Sellar et al., 2019 |