During the past 40 years, the climate model development overseas has gone through several key stages, which are summarized as follows. The first stage is the rapid development of an Atmospheric General Circulation Model (AGCM) in the mid-1970s. The second stage is the coupling of the AGCM with a land model in the mid-1980s, while sea surface temperature and sea ice were prescribed. The third stage is the coupling of the AGCM with a “slab” ocean in the early 1990, which was used to run climate projection experiments in the First Assessment Report (FAR) of the Intergovernmental Pa-nel on Climate Change (IPCC).
In the fourth stage, which occurred in approximately 1995, the AGCM began to include sulfate aerosol and volcanic aerosol and started coupling with an Oceanic General Circulation Model (OGCM). A fully coupled system including the atmosphere, ocean, land, and sea ice was used for climate projection experiments in the Second Assessment Report (SAR) of the IPCC.
In the fifth stage, which occurred in approximately 2001, the runoff process was added to the land model to close the hydrological cycle. In this period, the AGCM could better deal with the aerosols and the OGCM could describe the Meridional Overturning Circulation (MOC). Many models began to take the carbon cycle in the climate system into account. The physical processes in the models supporting the Third Assessment Report (TAR) of the IPCC were more comprehensive.
In the sixth stage, which occurred in approximately 2007, dynamic vegetation was first introduced into the land model, and the AGCM began to include atmosphe-ric chemical processes. These new models were used for climate projection experiments in the Fourth Assessment Report (AR4) of the IPCC. At this stage, an important advance was that the “flux correction” technique was no longer employed by most models.
The seventh stage occurred in approximately 2013. In this stage, the CSM was gradually evolving into the ESM. A comprehensive improvement was observed in terms of simulating the biogeochemical cycle in the oceanic ecosystem module and addition of the global vegetation dynamic module in the land model.
At each stage above, the model resolution was also becoming finer with advances in high-performance computing. Especially in the latest five years, the atmosphe-ric models have started to gradually evolve to the convection-resolving scale and the ocean models have started to gradually evolve to the eddy-resolving scale.
On the global scale, the World Climate Research Program (WCRP) plays a pivotal role in driving internatio-nal cooperation in climate model development and application. Initiated in 1995 by the Working group on Coupled Modelling (WGCM) affiliated with the WCRP, the Coupled Model Intercomparison Project (CMIP) is now in its sixth phase. The CMIP plays an irreplaceable role in setting the standards for climate model experiments and validations, unifying data formats, and normalizing international data-sharing mechanisms. The CMIP data have been widely used in international climate research, which supports the compiling of the IPCC assessment reports and climate change negotiations.
Climate model development in China began in the 1980s, nearly at the same time as developed countries. After a more than 30-yr endeavor, remarkable achievements have been made in the AGCM, OGCM, land-process model, coupled model, and so on (see the review article of Wang et al., 2008). Corresponding to the mo-del development, R&D teams in China are growing increasingly larger. Taking the Chinese models involvement in every IPCC scientific assessment report and CMIP activity as an example (Table 1), the evolution of CSM R&D pattern in China can be clearly observed. In the 1992 IPCC Supplement to the FAR, a Chinese model was included that involves a two-layer atmospheric mo-del coupled with a mixed-layer ocean model developed by the Institute of Atmospheric Physics, Chinese Academy of Sciences (IAP/CAS) (Wang et al., 1993). In the IPCC SAR published in 1995, the model from IAP/CAS was updated to a 20-layer air–sea coupled model, in which a thermodynamic sea-ice model was added in the OGCM and the “flux correction” technique was used in the air–sea coupling (Zhang et al., 1992; Chen et al., 1997a, b). Thereafter, a Global Ocean–Atmosphere–Land System model (GOALS) was established based on a 20-layer OGCM and a 9-layer spectral AGCM. The second version, GOALS2, participated in the First Phase of the CMIP (CMIP1; Wu et al., 1997; Yu and Zhang, 1998). In 2001, the IPCC TAR cited the results from the CMIP2 models, to which GOALS4, which was developed by the IAP/CAS, contributed. Compared with GOALS2, GOALS4 had an improved diurnal cycle of solar radiation (Shao et al., 1998) and freshwater flux exchange process in air–sea coupling (Zhou et al., 2001). Due to a major improvement in GOALS, Flexible GOALS (FGOALS) was developed, which first realized direct coupling in China without the “flux correction” based on the coupler framework (Yu et al., 2004). The first version of FGOALS, i.e., FGOALS-g1.0, participated in the CMIP3, which was cited by the IPCC AR4 in 2007. During nearly 20 years from the FAR to AR4, the CSM from CAS supported China’s entry into the international competition in climate modeling and projection. From the CMIP1 to CMIP3, the number of participating international models increased from 10 to 18 and then to 23, in which the CSM from CAS was the only model from a developing country (see the review article by Zhou et al., 2014). This R&D pattern in China was dramatically changed in the CMIP5.
IPCC CMIP Model Horizontal resolution of atmospheric model Horizontal resolution of oceanic model FAR — IAP 4° × 5°, L2 Mixed layer ocean model SAR CMIP1 GOALS2 4° × 5°, L2 4° × 5°, L20 TAR CMIP2 GOALS4 R15 (4.5° × 7.5°), L9 4° × 5°, L20 AR4 CMIP3 FGOALS-g1.0 2.8° × 2.8°, L26 1.0° × 1.0°, L30 *FAR: First Assessment Report; SAR: Second Assessment Report; TAR: Third Assessment Report; and AR4: Fourth Assessment Report
Table 1. The models from China used for IPCC AR1–4
The CMIP5 results were cited by the IPCC AR5 in 2013. There were 35 models worldwide that participated in the CMIP5, among which 6 models were from China (Table 2). They are as follows: FGOALS-g2 and FGOALS-s2, which are from the IAP/CAS; BCC-CSM and BCC-CSM1.1m, which are from the Beijing Climate Center and include the carbon cycle (Wu et al., 2014); BNU-ESM from the Beijing Normal University; and FIO-ESM from the First Institute of Oceanography under the Ministry of Natural Resources (Former State Oceanic Administration) of China. Both BNU-ESM and FIO-ESM were developed based on the third version of the Community Climate System Model (CCSM3; Collins et al., 2006). BNU-ESM updated the OGCM in CCSM3 and introduced an independently developed land model, while FIO-ESM first brought small-scale waves into CSM through a vertical mixing process to represent the controlling effect of waves to the upper ocean and climate (see the review article of Zhou et al., 2014). The increasing number of Chinese models in the CMIP reflects the expansion of the R&D teams focusing on modeling. Since the CMIP5, the CSM R&D pattern in China has transformed from the dominant status of the CAS into a tripartite situation involving the CAS, ministries, and universities. This situation will be reinforced in the CMIP6 (Zhou et al., 2019).
Model Institute Horizontal resolution of atmospheric model Horizontal resolution of oceanic model BCC_CSM1.1 BCC 2.8° × 2.8° 0.8° × 1.0° BCC_CSM1.1(m) BCC 1.1° × 1.1° 0.8° × 1.0° BNU-ESM BNU 2.8° × 2.8° 0.9° × 1.0° FGOALS-g2 LASG-CESS 3.0° × 2.8° 0.9° × 1.0° FGOALS-s2 LASG 1.7° × 2.8° 0.9° × 1.0° FIO-ESM FIO 2.8° × 2.8° 0.9° × 1.0°
Table 2. The models from China that participated in CMIP5 and used in IPCC AR5
The Coupled Model Intercomparison Project Phase 6 (CMIP6) was planned in 2013, officially launched in 2016, and is currently being implemented. The CMIP6 will directly support the preparation of the IPCC Sixth Assessment Report. Globally, there are 33 model R&D teams participating in the CMIP6, with 13 new teams that did not participate in the CMIP5. There are 112 model versions① registered by these 33 institutions participating in the CMIP6. Ten model teams in China have signed up for the CMIP6. Compared with the CMIP5, the model teams that are independently participating in the CMIP for the first time include the ESM teams from the Chinese Academy of Sciences, the Chinese Academy of Meteorological Sciences (CAMS), Nanjing University of Information Science & Technology (NUIST), Tsinghua University (THU), and the Research Center for Environmental Changes (RCEC), Taiwan, China (Zhou et al., 2019).
Although the number of climate models participating in the CMIP6 in mainland China has increased significantly compared with that in the CMIP5, the basic R&D team pattern, i.e., the three-component configuration involving the Chinese Academy of Sciences, ministries, and universities, remains unchanged. There are three model versions from the Chinese Academy of Sciences, namely, CAS FGOALS-g3, CAS FGOALS-f3, and CAS-ESM; the three models from the ministries are BCC-CSM (and BCC-ESM) from the National/Beijing Climate Center, China Meteorological Administration, CAMS-CSM from the Chinese Academy of Meteorological Sciences, and FIO-ESMv2.0 from the First Institute of Oceanography, Ministry of Natural Resources (MNR); the three models from the universities include BNU-ESM-1-1 from Beijing Normal University, CIESM from Tsinghua University, and NESM v3 from Nanjing University of Information Science & Technology. The active participation of universities in the R&D of CSMs is a new trend in the past five years, which is important for the cultivation of R&D talents for model development. The main technical features of the Chinese models participating in CMIP6 are summarized as follows. The main technical metrics are shown in Table 3. The table shows that three of the ECSMs are based on the Community Earth System Model (CESM) from the NCAR of the U.S. In addition, the ocean component models of the three ECSMs are based on the Modular Ocean Model (MOM) from the Geophysical Fluid Dynamics Laboratory (GFDL) of the U.S. It is noted that the model from Beijing Normal University is the same as that in CMIP5 (Ji et al., 2014), which was introduced in the previous review on Chinese contributions to the CMIP5 (Zhou et al., 2014) and will not be repeated here. Below is an overview of the remaining model versions.
Model Institute Atmospheric model Oceanic model Land surface model Sea ice model Coupler 1 BCC-CSM2-MR
(Wu et al., 2019a, b;
Xin et al., 2019)
(T42, ~280 km, L26)
(T106, ~120 km, L46)
(T266, 45 km, L56)
BCC-AVIM2 SIS CPL5 2 BNU-ESM-1-1
(Ji et al., 2014)
(2.8°[~280 km], L26)
MOM4p1 (1.0° at high lati-
tudes, close to 0.3° in the tropics, L40)
CoLM CICE4 CPL 3 CAMS-CSM
(Rong et al., 2019)
(T106, ~120 km, L31)
MOM4 (1.0° at high lati-
tudes, close to 0.3° in the tropics, L50)
CoLM SIS FMS-coupler 4 CAS-ESM ICCES, IAP, CAS IAP AGCM5
(1.4°[~140 km], L35)
LICOM2 (1.0°, 0.5° in the meridional directions over the tropics, L30) CoLM CICE4 CPL7 5 CAS FGOALS-f3
(Bao et al., 2019)
LASG, IAP, CAS FAMIL2.2
(100 km, 25 km, L32)
LICOM3 (1°/L30, 0.1°/L55) CLM4 CICE4 CPL7 6 CAS FGOALS-g3
(Tang et al., 2019)
LASG, IAP, CAS GAMIL3
(2°[~200 km], L26)
LICOM3 (1.0°, 0.5° in the meridional directions over the tropics, L30) CAS-LSM CICE4 CPL7 7 CIESM
(Lin et al., 2019)
CESS, THU Modified CAM5
(~100 km, L30)
POP2 (1°, L60) CLM4 CICE4 C-coupler2 8 FIO-ESM2.0
(Song et al., 2019)
FIO, NMR CAM5
(~100 km, L30)
POP2 (1.1° at high lati-
tudes, 0.3°–0.5° in the tropics, L61)
CLM4 CICE4 CPL7 9 NESM v3
(Cao et al., 2019)
(T63, ~200 km, L47)
NEMO v3.4 (1.0° at high latitudes, close to 0.3° in the tropics, L46) JABACH CICE4 OASIS3-MCT 10 TaiESM① RCEC, Taiwan, China CAM5.3
(~100 km, ~200 km, L30)
POP2 (1.0°, L70) CLM4 CICE4 CPL7 Note: Most acronyms can be found in the main text. ICCES: International Center for climate and Environment Sciences; CESS: Center for Earth System Science. ①Overview of Chinese contribution to CMIP6: Status and challenges, Presentation by T. J. Zhou at Pan-WCRP modelling Meeting, Met Ofﬁce, Exeter, 9–13 October 2017.
Table 3. Metrics of current Earth/Climate System Models that have participated in CMIP6 from China
There are three new models developed by the Natio-nal (Beijing) Climate Center that participated in the CMIP6, including BCC-ESM1.0, medium-resolution climate model BCC-CSM2-MR, and high-resolution climate model BCC-CSM2-HR. The ice model component is the Sea Ice Simulator (SIS) developed by GFDL, and the land surface model for each model is BCC-AVIM2.0. BCC-AVIM2.0 was developed from the earlier version BCC-AVIM1.0, in which the parameterization scheme was improved to calculate the threshold temperature to initiate the freezing (thawing) of soil water (ice), and included the parameterization of snow surface albedo as well as the description of rice paddies in the vegetation module, and so on (Li W. P. et al., 2019). The ocean model is MOM4-L40 in BCC-ESM1.0 and BCC-CSM2-MR, which use a tripolar grid with a resolution of 0.3° in the tropics, increased to 1.0° in the polar region. In BCC-CSM2-HR, the ocean model is MOM5-L50 with a resolution of 0.25°. There is large discrepancy between the atmospheric models of BCC-ESM1.0, BCC-CSM2-MR, and BCC-CSM2-HR. In BCC-ESM1.0, the atmospheric model is BCC-AGCM3-chem, which has a horizontal resolution of T42 (~280 km) and 26 vertical layers. The top of the model is at 2.917 hPa. The chemistry module and aerosol process are included in BCC-AGCM3-chem. The atmospheric model of BCC-CSM2-MR is BCC-AGCM3-MR, which has a horizontal resolution of T106 (~120 km) and 46 layers in the vertical direction, with the model top at 1.450 hPa. In BCC-AGCM3-MR, the schemes of the atmospheric radiation, deep convection process, and the gravity wave drag were improved, and the indirect effects of aerosols were included (Wu et al., 2019a, b). The atmospheric model of BCC-CSM2-HR is BCC-AGCM3-HR, which has a horizontal resolution of 45 km and 56 layers in the vertical direction, with the model top at 0.1 hPa. The details of these models can be found in Xin et al. (2019).
The CAMS-CSM has an atmospheric component based on the ECHAM5 (v5.4) from the Max Planck Institute for Meteorology (MPI-M). The horizontal resolution of the model is T106 (~120 km) with 31 vertical layers extending from the surface to 10 hPa. Compared with ECHAM5, major improvements of the CAMS-CSM atmospheric model include the following: (1) a Two-step Shape Preserving Advection Scheme (TSPAS) is used for water vapor transport; (2) a relevant k-distribution scheme (BCC_RAD) developed by Beijing Climate Center is adopted for radiative transfer parameterization. The ocean component of the CAMS-CSM is MOM4 from GFDL. The MOM4 component uses a tripolar grid with a zonal resolution of 1° globally and a meridional resolution of 1/3° within 10°S–10°N, and it changes to 1° poleward of 30°S (N). There are 50 vertical layers, and the top 23 layers have an even resolution of 10 m. The sea ice component is the SIS from GFDL. The land surface component is the Common Land Model (CoLM) and the unfrozen water below 0°C in soil is considered in the CoLM model. The details of CAMS-CSM can be found in Rong et al. (2019).
The ESM developed by the CAS is named CAS-ESM. The atmospheric component model of CAS-ESM is the IAP AGCM5, which has a horizontal resolution 1.4° (~140 km). Compared to the previous version, i.e., IAP AGCM4 (Zhang et al., 2013), the number of vertical levels has increased from 30 to 35, and a new flexible leaping-point scheme is designed to replace the filter. A new deep convection scheme and a new cloud macrophysical scheme are designed to improve the precipitation simulation. Some modifications in the cloud microphysics scheme (Xie and Zhang, 2015) and the boundary layer turbulence scheme are implemented. A new orographic gravity waves drag scheme considering the subgrid topography is also developed. The ocean component model LICOM2, which has a horizontal resolution of 1.0° and a refined meridional resolution of 0.5° near the equator, designs a well-posed salinity boundary condition that considers the real salt flux and introduces the Fairall method as a new way to calculate air–sea turbulence fluxes. The land component model is the CoLM, in which a multi-layer soil organic carbon scheme and a three-dimensional (3-D) vegetation canopy radiative transfer scheme are developed to improve the permafrost carbon feedback and land surface energy budget. The sea ice component model is the Los Alamos National Laboratory (LANL) sea ice model version 4.0 (CICE4). The coupler component is CPL7, with modifications to support 3-D variable interpolation and communication.
The CAS FGOALS-f3 developed by the IAP/CAS includes four component models (atmosphere, ocean, land, and sea ice) and a coupler. In terms of horizontal resolution, the CAS FGOALS-f3 model has two versions, i.e., low resolution (FGOALS-f3-L) and high resolution (FGOALS-f3-H) (Bao et al., 2019; He B. et al., 2019; Li J. X. et al., 2019). The atmospheric component models of both model versions are the Finite-volume Atmospheric Model of IAP/LASG (FAMIL) version 2.2, which uses a finite volume cubic spherical grid (Zhou et al., 2015). The horizontal resolution of the atmospheric model in FGOALS-f3-L is approximately 1° (~100 km), while it is approximately 0.25° (~25 km) in FGOALS-f3-H. There are 32 levels in the vertical direction, and a mixed coordinate system with a mode top of 2.16 hPa (approximately 45 km) is used. For the physical process parameterization, FAMIL2.2 uses an explicit convective precipitation scheme developed by IAP-LASG, which improves the simulations of tropical cyclones, Madden–Julian Oscillation, and extreme precipitation (He B. et al., 2019; He S. C. et al., 2019; Li J. X. et al., 2019). The land surface model of CAS FGOALS-f3 is the Community Land Model version 4.0 (CLM4). The ocean model component is LICOM3, which is the same as that in the CAS FGOALS-g3. In FGOALS-f3-L, the ocean model has a horizontal resolution of approximately 1° and 30 levels in the vertical direction, while in FGOALS-f3-H, a horizontal resolution of 0.1° and 55 layers in the vertical direction are used. The sea ice component of CAS FGOALS-f3 is CICE4 and the coupler is CPL7.
The CAS FGOALS-g3 is also developed by the IAP/CAS. The atmospheric model component is GAMIL3. Compared with the previous version, i.e., GAMIL2, the horizontal resolution is increased from 2.8° to 2° (~200 km), and the water vapor advection scheme is further modified to strictly guarantee the water vapor conservation in the polar region. A two-dimensional instead of a one-dimensional Message Passing Interface, with both zonal- and meridional-split flows, is introduced (Liu L. et al., 2014). The improvements of the physics parameterizations include the following: inclusion of the convective momentum transport and stratospheric aerosols (Nie et al., 2019), a stratiform cloud scheme based on the estimated inversion temperature (Guo and Zhou, 2014), a new stratus cloud amount scheme, a boundary layer scheme based on turbulent flow energy, and an aerosol module developed by the MPI-M (MACv2-SP). The ocean component is LICOM3 with a tripolar grid. The horizontal resolution is 1°, which is increased to 0.5° at the equator. The GM (Gent–McWilliams) vertical mixing and tidal mixing are improved. The land surface model is CAS-LSM, which was developed based on CLM4.5 by coupling with a scheme that considers the lateral flow of groundwater, the use of water by human activities, changes in the soil freezing and melting interface, and the effects of river nitrogen transport processes. The sea ice is CICE4 and the coupler is CPL7. The details can be found in Tang et al. (2019).
Based on the NCAR Community Earth System Model (CESM1.2.1), Tsinghua University, in collaboration with several institutes in China, developed the Community Integrated Earth System Model (CIESM1.1). The atmospheric component is based on the Community Atmospheric Model version 5 (CAM5), which has a horizon-tal grid spacing of ~100 km and 30 vertical levels with the model top at 1 hPa. The modifications to the atmospheric model include a modified deep convection scheme, a four-stream shortwave radiation scheme, a diagnostic statistical cloud macrophysics scheme, a single-ice microphysics scheme, and prescribed aerosol radiative effects. These modifications alleviate some long-lasting systematic biases in the CESM, such as the underestimated marine boundary layer clouds and the double ITCZ problem. The ocean model is based on the LANL Parallel Ocean Program version 2 (POP2) and has a nominal grid of 1° with 60 vertical layers. Modifications to POP2 include a new ocean grid, a more efficient barotropic solver, modified mixing schemes, and a new air–sea flux exchange method. The land model is based on CLM4, with use of new soil texture and organic matter content data and a new thermal roughness parameterization. The sea ice model is CICE4.1, with the same grid structure as the ocean model. A few modifications related to the physics and dynamics of sea ice, such as the salinity-dependent freezing temperature and the floe-size dependent lateral melting rates are introduced. The Community Coupler version 2 (C-Coupler2) is employed in the CIESM to handle various model coupling functions, such as data transfer and data interpolation, among others. All the testing and experiments were conducted on the “TaihuLight” supercomputer at Wuxi, China. More details can be found in Lin et al. (2019).
The FIO-ESM v2.0, developed by the FIO, MNR of China, contains the following five model components: the atmospheric model of CAM5, the land surface model of CLM4, the sea ice model of CICE4, and the ocean model of the modified POP2 incorporating the MArine Science and NUmerical Modeling (MASNUM) surface wave model. The horizontal resolutions of CAM5 (with 30 vertical layers) and CLM4 are approximately 100 km, and a nominal 1° is applied for POP2, the MASNUM surface wave model, and CICE4. Four distinctive physi-cal processes are included, which are the surface wave-induced vertical mixing, air–sea flux modulated by stokes drift, heat flux associated with sea spray, and SST diurnal cycle. Further information on FIO-ESM v2.0 can be found in Song et al. (2019).
The 3rd version of the NUIST Earth System model (NESM v3) has been developed in recent years and is a newcomer to the CMIP6. The NESM v3 consists of the ECHAM v6.3 atmospheric model, which is directly coupled with the JSBACH land surface model, the Nucleus for European Modelling of the Ocean (NEMO) model v3.4, the CICE v4.1 sea ice model, and the OASIS3-MCT_3.0 coupler. The ECHAM v6.3 and JSBACH are adopted from the Max Planck Institute ECHAM se-rial model. The horizontal resolution of the atmospheric component model is T63, which corresponds to approximately 200 km in the meridional and zonal directions. The 47 vertical layers extend from the surface up to 0.01 hPa. Improvements in the atmospheric component mo-del include the development of cumulus parameterization, macro-/microphysics, and a boundary layer scheme. The horizontal resolution of the land surface model is the same as that of the ECHAM v6.3, but with 5-layer soil. The ocean component model is NEMO v3.4 with a horizontal resolution of approximately 1° × 1°. The resolution in the meridional direction is refined to 1/3° over the tropical region. There are 46 layers in the vertical direction, with the first 15 layers in the top 100 m. A “stiffer” oceanic equation of the state and the mix layer brine rejection process are incorporated in the ocean model. The sea ice component model is CICE v4.1, which was originally developed at LANL. Its resolution is approximately 1° × 1/2° in the meridional and zonal directions, with four sea ice layers and one snow layer on the top of the ice surface. Both the sea ice and snow albedo schemes are improved in the sea ice component model. Further information on NESM v3 can be found in Cao et al. (2019).
The ESM developed by the Academia Sinica, Taiwan, China, is named TaiESM. The model is an improved version of CESM1.2.2. The atmospheric model is CAM5.3, which has two horizontal resolutions of ~100 and ~200 km and 30 layers in the vertical direction. The ocean model is POP2, with a horizontal resolution of 1.0° and 70 layers in the vertical direction. The land surface mo-del is CLM4, and the sea ice model is CICE4. The land surface, ocean, and sea ice components of TaiESM remain unchanged compared to CESM version 1.2.2. The main improvements are the deep convective parameterization scheme, the cloud macroscopic physical process, the aerosol module, and the inclusion of the effects of the topography on solar radiation.
In the past five years, in addition to the great achievement in the R&D of ECSMs in China, considerable progress has also been made in the construction of the R&D platform of ECSMs. The National 12th Five-Year Major Science and Technology Infrastructure “Earth System Science Numerical Simulator Facility,” named “EarthLab,” has undergone project application, feasibility research, preliminary program planning, and budgetary calculation, and it was officially started in the autumn of 2018 with approval by the National Development and Reform Commission of China. The facility was built by the IAP/CAS and Tsinghua University. The facility includes five major systems, i.e., earth system numerical simulation software, regional high resolution environmental simulation software, technical support and management for super computing, supported database and data assimilation and visualization, and high performance computers for earth science. The total investment is more than 1.2 billion RMB, and the construction period is 4 yr.
The scientific goals of the EarthLab are to deeply understand the basic laws of the complex system of the earth’s environment; to explore the physical, chemical, and biological processes of the earth’s atmosphere, hydrosphere, cryosphere, biosphere, and lithosphere; to explore the impacts of each sphere and the associated interactions on the earth system and the regional environment of China; to fuse the simulation and the observation data to improve the prediction accuracy; to achieve quantitative characterization and simulation of the complex process of the earth system at the mesoscale; to provide scientific support for major issues, such as national disaster prevention and mitigation, climate change, and atmospheric environmental governance; to promote interdisciplinary integration and fusion between different disciplines of Earth System Science; and to promote the advancement of China’s Earth System Science to the world-class level.
The hardware construction target of the EarthLab is a dedicated structure of a high-performance computer system with a peak value of not less than 15 Petaflops and a storage of not less than 80 PB. The overall performance will be significantly higher than other high-performance computers currently dedicated to earth system numerical simulations. The proposed ESM includes the subsystems of the atmosphere, ocean, land surface, aerosol and atmospheric chemistry, terrestrial vegetation ecosystem, marine biochemical system, terrestrial biochemical system, and the associated coupled models; and it will have complete, self-consistent, and rigorous dynamic, physical, chemical, and ecological processes, with Chinese independent intellectual property and original characteristics. The resolution will reach a globally advanced level. The ESM will have a horizontal resolution of approximately 10–25 km, which will significantly improve the model’s ability to simulate past and future climate changes. Moreover, we will develop a high resolution regional environmental model with a spatial resolution of 1–3 km for high resolution simulations of weather, air pollution, long-term climate change risks, and agricultural drought at the cloud resolving scale in China and in the surrounding areas.
3.1. Chinese ECSMs participating in CMIP6
3.2. The National 12th Five-Year Major Science and Technology Infrastructure “Earth System Science Numerical Simulator Facility”
Up to now, four Chinese model groups, i.e., BCC-CSM2-MR, CAMS-CSM1.0, CAS FGOALS-g3, and NESM3, have officially published their model results of some CMIP6 experiments to the Earth System Grid Federation (ESGF) through the Earth System Grid (ESG) node. To demonstrate the basic performance of the Chinese models that participated in the CMIP6, we will show the ensemble mean results from these four model in terms of climatology, historical climate changes in the 20th century, and future projections. The third realization from BCC-CSM2-MR is used, while the first realization from the other three models is used.
Figure 1 shows the observed and simulated December–January–February (DJF) means and the June–July–August (JJA) mean surface air temperature (SAT) during 1995–2014 and the corresponding model biases. The ensemble mean of the four Chinese models well captures the spatial distributions of observed SAT in both boreal summer and winter, but tends to overestimate the SAT over most oceans, the Northeast Eurasian continent in boreal winter, and the center of Eurasia in boreal summer (0.50 to 4.00°C); and it tends to underestimate the SAT over most land regions and the Arctic (–0.50 to –4.00°C). The evident cold biases in the high latitude region of the Northern Hemisphere in boreal winter may be associated with the overestimated sea ice. The performance of the ensemble mean is better than that of the individual model (Table 4).
Figure 1. 1995–2014 (a) DJF and (c) JJA mean surface air temperature (°C) derived from the observation (Obs, shading) and the multi-model ensemble mean (MME, contour) of four Chinese models that participated in CMIP6. (b) DJF and JJA mean biases of the MME, in which the dotted areas denote values passing the 95% confidence level. The observation is the average of HadCRUT4 (Hadley Centre and Climate Research Unit Temperature) and BEST (Berkeley Earth Surface Temperatures) data.
Model Temperature Precipitation PCC RMSE PCC RMSE JJA DJF JJA DJF JJA DJF JJA DJF MME 0.99 0.99 1.65 2.11 0.84 0.85 1.44 1.42 BCC-CSM2 0.99 0.99 2.83 2.47 0.82 0.82 1.66 1.74 CAMS-CSM 0.99 0.98 2.10 3.18 0.74 0.76 1.93 1.74 FGOALS-g3 0.99 0.99 2.17 2.96 0.71 0.81 2.10 1.69 NESM3 0.99 0.99 1.92 2.35 0.80 0.76 1.75 1.89
Table 4. Pattern correlation coefficients (PCCs) and root mean square errors (RMSEs) of surface air temperature (°C) and precipitation (mm day−1) between the observation and the historical simulations derived from four Chinese models and the multi-model ensemble mean
The observed and simulated DJF mean and JJA mean precipitation during 1995–2014 and the corresponding model biases are shown in Fig. 2. The precipitation in the observation ranges from 0.20 to 10.00 mm day−1, with the center in the tropical ocean (7.00–10.00 mm day−1) and global monsoon regions in summer (~7.00 mm day−1). The ensemble mean of the four Chinese models well reproduces the observed patterns (Table 4). Some biases are also obvious. Wet biases are seen over the ITCZ regions, the Australian monsoon region, the southern South Asian monsoon region in boreal summer, East Asia in boreal winter, the western coast of North America, and most regions in the Southern Hemisphere in boreal winter (0.50–4.00 mm day−1), while dry biases are seen over the northern Eurasian continent, East Asia, southern Asia, South America, Southern Hemisphere in boreal summer, and southern Atlantic Ocean in boreal winter (~2.00 mm day−1). In the tropics, the “double ITCZ” model bias is evident. The “double ITCZ” bias commonly exists in other CMIP3 and CMIP5 models (Lin, 2007; Li and Xie, 2014). Previous studies suggest that this kind of model bias may be associated with a too-strong Bjerknes feedback and SST–latent heat flux feedback but a too-weak SST–shortwave feedback (Lin, 2007), or the biases in the SST initialization over the center of the Pacific (Liu et al., 2012), or the inaccurate atmospheric model simulations of clouds (Li and Xie, 2014).
Figure 2. As in Fig. 1, but for precipitation (mm day−1). The observed precipitation is derived from GPCP data.
To compare the performance of the Chinese models with other CMIP6 models, we show the root mean square errors (RMSEs) of simulated global SAT and precipitation from the 4 Chinese models and 16 other CMIP6 models (Fig. 3). Compared with the other CMIP6 models, the Chinese models have relatively large biases in the simulation of precipitation in boreal winter. In summer, the global SAT and precipitation simulated by the Chinese models are comparable to the other CMIP6 models in terms of the RMSE. Spatially, the biases in the Chinese models are also shown in the other models (figures omitted). In addition, the biases in the ensemble mean of the four Chinese models are smaller than those of most of the CMIP6 models, and the performance of the Chinese model’s ensemble mean is close to the performance of ensemble means of 20 CMIP6 models.
Figure 3. (a) DJF and (b) JJA mean root mean square error (RMSE) of simulated global surface air temperature (°C) and precipitation (mm day−1) from four Chinese models (red dot) and other 16 models (gray dot) that participated in CMIP6. The RMSE of the four Chinese models ensemble mean (red triangle) and 20 CMIP6 models ensemble mean (black triangle) are also presented.
China is mainly affected by the Asian monsoon. Therefore, the model’s monsoon simulation is always a focus of Chinese model R&D teams. Figure 4 shows the spatial distributions of JJA mean wind fields at 850 hPa and the corresponding biases relative to the Japanese 55-yr Reanalysis (JRA55). The Chinese models well reproduce the spatial distribution of the observed Asian summer monsoon circulation (PCC = 0.96). Some biases are also evident. The simulated Northwest Pacific Subtropi-cal high (NPSH) and South Asian summer monsoon are weaker than the observed. The location and strength of the NPSH determine the water vapor transport for the East Asian summer rain belt, and such biases could directly influence the simulation of monsoon rainfall.
Figure 4. Spatial patterns of the June–July–August–September (JJAS) mean low-level wind field (m s−1) at 850 hPa averaged for 1995–2014 derived from (a) JRA55 and (c) multi-model ensemble mean of historical simulations from the four Chinese models that participated in CMIP6. (b) The differences between ERA-Interim and JRA55. (d) The simulated bias of the Chinese models, relative to the mean of JRA55 and ERA-Interim data.
Historical temperature change in the 20th century is a fundamental metric for evaluating the model’s performance. Figure 5 shows the series of annual SAT anomalies between 60°S and 60°N from the observation and models. In observation, global mean temperature increased by 0.74 [0.71–0.79]°C from 1900 to 2014. The warming of 4 Chinese models ensemble mean is 0.73°C, with a range of [0.63–0.91]°C, which is close to the observation. The strongest warming is found in FGOALS-g3 (0.91°C), while the weakest is 0.63°C in CAMS-CSM1.0, and the warming in NESM3 (0.68°C) and BCC-CSM2-MR (0.71°C) are in between. In CAS FGOALS-g3, the change of external forcing (greenhouse gas and aerosol) in CMIP6 is one of the main factors that induce stronger warming trend compared to that in CMIP5 (Nie et al., 2019). An increasing trend is also seen in observed global land precipitation but obvious difference between observation and models are found. One of the reasons is that observational rainfall datasets have large uncertainties, especially in the early 20th century. Moreover, the changes of precipitation exhibit evident regional features, which are strongly influenced by the internal variability.
Figure 5. Anomalies of (a) annual mean surface air temperature (MST, °C) between 60°S and 60°N and (b) global land averaged rainfall (%, divided by the 1900–1950 mean) from the observation and the multi-model ensemble mean of historical simulations from four Chinese models. A 20-yr running average was applied to the precipitation anomaly. The surface air temperature (global land averaged rainfall) anomalies were calculated as departures from the annual mean of 1850–1900 (1900–1950).
The tier-1 experiments of projections in CMIP6 include the projections under the Shared Socioeconomic Pathway (SSP) 1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5. Figure 6 shows the projected changes of the global SAT and land precipitation from these four scenarios. In the near-term (2021–2040) projection, the temperature warming (0.55–0.71°C) and precipitation change (0.76%–1.63%) are similar among the four scenarios; while the differences of projected SAT changes are evident in the mid-term (2041–2060), but the differences of projected precipitation changes are still not evident. In long-term (2081–2100) projection, there are significant differences among the four scenarios in the projected SAT warming (0.72–3.13°C); in contrast, the differences of projected precipitation changes among different scenarios are not so evident as in temperature changes and show large uncertainty.
Figure 6. Projected changes of (a) global mean surface air temperature (GMST, °C) and (b) percentage changes of global land averaged rainfall (%) with respect to the period 1995–2014 under Shared Socioeconomic Pathway (SSP) 1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5 scenarios derived from four Chinese models. A 20-yr running average was applied to the anomalies. The data before 2014 are from the historical simulations. Solid lines denote the multi-model ensemble mean of four models, while shadings denote the uncertainty range of 5th to 95th percentile.
Figure 7 shows spatial patterns of the long-term (2081–2100) projected SAT changes under SSP1-2.6 and SSP5-8.5 scenarios. The largest warming is seen in high latitude regions of the Northern Hemisphere. The warming over land is larger than that over ocean. The warming amplitude is rather different between the two scena-rios. The SAT warming in high latitude regions of the Northern Hemisphere exceeds 7.00°C in SSP5-8.5, while it is just 1.00–2.00°C in SSP1-2.6. Moreover, there are also large differences in the projected SAT changes among the four models (Table 5). In SSP1-2.6, the warming in NESM3 is strongest (1.18 ± 0.13°C), while that in FGOALS-g3 is weakest (0.48 ± 0.10°C), and the warming in BCC-CSM2 (0.72 ± 0.09°C) and that in CAMS-CSM (0.67 ± 0.16°C) are within this range. In SSP5-8.5, the warming in NESM3 is also strongest (4.27 ± 0.37°C), while that in CAMS-CSM1.0 (2.52 ± 0.20) is weakest, and the warming in BCC-CSM2 (3.13 ± 0.23°C) and that in FGOALS-g3 (2.75 ± 0.23°C) are within this range.
Model and period SSP1-2.6 (°C) SSP2-4.5 (°C) SSP3-7.0 (°C) SSP5-8.5 (°C) MME 2021–2040 0.55 ± 0.22 (0.18, 0.87) 0.59 ± 0.21 (0.24, 0.93) 0.58 ± 0.09 (0.42, 0.73) 0.71 ± 0.23 (0.33, 1.08) 2041–2060 0.75 ± 0.33 (0.21, 1.29) 1.05 ± 0.26 (0.62, 1.47) 1.14 ± 0.19 (0.82, 1.45) 1.46 ± 0.43 (0.76, 2.16) 2081–2100 0.76 ± 0.30 (0.28, 1.25) 1.61 ± 0.37 (1.00, 2.22) 2.43 ± 0.38 (1.81, 3.04) 3.17 ± 0.78 (1.89, 4.44) BCC-CSM2-MR: 2081–2100 0.72 ± 0.09 (0.57, 0.87) 1.63 ± 0.13 (1.42,1.84) 2.84 ± 0.23 (2.46, 3.21) 3.13 ± 0.23 (2.75, 3.51) CAMS-CSM1.0: 2081–2100 0.67 ± 0.16 (0.42, 0.93) 1.31 ± 0.12 (1.12, 1.51) 2.10 ± 0.21 (1.75, 2.44) 2.52 ± 0.20 (2.20, 2.84) FGOALS-g3: 2081–2100 0.48 ± 0.10 (0.32, 0.64) 1.36 ± 0.10 (1.19, 1.53) 2.34 ± 0.26 (1.92, 2.77) 2.75 ± 0.23 (2.38, 3.12) NESM3: 2081–2100 1.18 ± 0.13 (0.97, 1.39) 2.13 ± 0.11 (1.94, 2.32) − 4.27 ± 0.37 (3.67, 4.87) Note: NESM3 does not provide the projection under SSP3-7.0 scenario.
Table 5. CMIP6 annual mean surface air temperature anomalies (°C) from the 1995–2014 reference period for selected time periods and Shared Socioeconomic Pathways (SSPs). The multi-model mean ± 1 standard deviation ranges across the individual models are listed and the 5%–95% ranges from the models’ distribution (based on a Gaussian assumption and obtained by multiplying the CMIP6 ensemble standard deviation by 1.64) are given in brackets. The values tabulated here are for single simulations from the four models that have thus far contributed to the CMIP6 exercise. The models are BCC-CSM2-MR, CAMS-CSM1.0, FGOALS-g3, and NESM3
Figure 7. Spatial patterns of the projected changes of surface air temperature (°C) in (a, b) DJF and (c, d) JJA under (a, c) SSP1-2.6 and (b, d) SSP5-8.5 scenarios for the period 2081–2100 relative to the period 1995–2014 by multi-model ensemble mean of four Chinese models that participated in CMIP6. The areas with dots indicate that the projected changes are with the same sign in at least three models.
Figure 8 presents the changes of long-term projected precipitation under SSP1-2.6 and SSP5-8.5 scenarios. The spatial patterns of projected changes are similar in the two scenarios, but the changes under SSP5-8.5 scenario are significantly larger than those in SSP1-2.6. In 2081–2100, precipitation would increase significantly over equatorial ocean and high latitude regions (0.50–3.00 mm day−1), while decrease significantly in the subtropics (−0.50 to −1.00 mm day−1) except for the subtropical ocean and Asian monsoon and North Africa monsoon regions in boreal summer.
Figure 8. Spatial patterns of the percentage changes of projected rainfall ( mm day−1) in (a, b) DJF and (c, d) JJA under (a, c) SSP1-2.6 and (b, d) SSP5-8.5 scenarios for the period 2081–2100 relative to climatological rainfall averaged for the period 1995–2014 by multi-model ensemble mean of four Chinese models that participated in CMIP6. The areas with dots indicate that the projected changes are with the same sign in four models.
How should we understand the significant disagreement in the projected climate changes across the different models under the same emission scenario? Actually, this disagreement is directly related to the differences in the distinct sensitivity that the models show to greenhouse gas emissions. One of the metrics to measure the model sensitivity is the Equilibrium Climate Sensitivity (ECS), which is the ultimate change in the global mean surface air temperature (GMST) under a doubled CO2 concentration relative to the pre-industrial level. ECS is a core index to represent the sensitivity of the model response to greenhouse gas emissions, which is closely related to the uncertainty in the long-term climate projection (Chen and Zhou, 2015; Zhou and Chen, 2015). The ECS values of 5 Chinese models in the CMIP6 are 3.03°C (Wu et al., 2019a), 2.27°C (Chen et al., 2019), 2.98°C, 2.84°C, and 4.65°C, respectively, for BCC-CSM2-MR, CAMS-CSM, FGOALS-f3, FGOALS-g3, and NESM3. Ensemble mean of the 5 models is 3.16 ± 0.89°C. Figure 9 shows the relation between projected long-term GMST changes (2081–2100) and ECS across 4 Chinese models (projection data of FGOALS-f3 under all scena-rios and NESM3 under SSP3-7.0 scenario unavailable currently). Higher model ECS corresponds to larger warming in the projection. Because greenhouse gas is the dominant agent of climate change under high emission scenario (e.g., SSP5-8.5), positive correlation between warming and ECS is more significant than that under low emission scenario (e.g., SSP1-2.6).