Quoted from: Li, Lijuan, Yongqiang Yu, Yanli Tang, Pengfei Lin, Jinbo Xie, Mirong Song, Li Dong et al. "The Flexible Global Ocean‐Atmosphere‐Land System Model Grid‐Point Version 3 (FGOALS‐g3): Description and Evaluation." Journal of Advances in Modeling Earth Systems 12, no. 9 (2020): e2019MS002012. https://doi.org/10.1029/2019MS002012 

      Since the late 1980s, considerable efforts have been made in developing, assessing, and improving atmospheric, oceanic, land, and sea ice models and coupled climate models, at the State Key Laboratory of Numerical Modeling for Atmospheric Sciences and Geophysical Fluid Dynamics (LASG), Institute of Atmospheric Physics (IAP), Chinese Academy of Sciences (CAS). To date, there have been six generations of coupled climate models (Bao et al., 2013; Chen et al., 1997; Li, Lin, et al., 2013; Wu et al., 1997; Yu & Zhang, 1998; Yu et al., 2002, 2004; Zhang et al., 1992) developed at LASG‐IAP. These coupled models have also contributed to each phase of the Coupled Model Intercomparison Project (CMIP) (Covey et al., 2003; Meehl et al., 2005) and assessment reports of Working Group I of the Intergovernmental Panel on Climate Change (IPCC) (IPCC, 1992, 1995, 2001, 2007, 2013).

      The latest generation of climate system models developed at LASG‐IAP is Version 3 of the Flexible Global Ocean‐Atmosphere‐Land System model (FGOALS3), which includes three parallel subversions (FGOALS‐g3, FGOALS‐f3‐L, and FGOALS‐f3‐H). All three subversions were established based on a similar coupling framework, in which atmospheric, oceanic, sea ice, and land component models are connected via a common flux coupler. The same oceanic and sea ice models are shared by these three subversions, but different atmospheric and land component models are used. The present study describes the basic configuration of the coupled model FGOALS‐g3 and evaluates its performance in terms of climatological means, climatic variability, and long‐term climate trend. The four component models of FGOALS‐g3 include Version 3 of the Grid‐Point Atmospheric Model of LASG‐IAP (GAMIL3) for the atmosphere, Version 3 of the LASG‐IAP Climate System Ocean Model (LICOM3) for the ocean, Version 4 of the Los Alamos sea ice model for sea ice (http://climate.lanl.gov/Models/CICE), and the CAS‐Land Surface Model (CAS‐LSM) for the land (Xie et al., 2018). According to the numerical experiment design for CMIP6 (Eyring et al., 2016), we have conducted lots of CMIP6 experiments, including the Diagnostic, Evaluation, and Characterization of Klima (DECK), historical simulations, Scenario Model Intercomparison Project (ScenarioMIP), Global Monsoons Model Intercomparison Project (GMMIP), and Ocean Model Intercomparison Project (OMIP), which are also published online in the Earth System Grid Federation (ESGF). And we are also conducting other MIPs, such as the Paleoclimate Modeling Intercomparison Project (PMIP) and Decadal Climate Prediction Project (DCPP). In the present study, DECK, historical, and ScenarioMIP simulations are analyzed and evaluated with emphasis on the mean state, and climate variability and change.

      GAMIL3 is updated from GAMIL2 (Li, Wang, et al., 2013). Both versions use the same finite difference dynamical core, which conserves many properties, such as total mass and effective energy under the standard stratification approximation (Wang et al., 2004) and employ a 26 vertical σ layers (pressure normalized by surface pressure) coordinate with the model top at 2.194 hPa. Compared with GAMIL2, GAMIL3 has many modifications with respect to parallel computing, horizontal resolution, water vapor advection scheme, physical processes, and external forcings. GAMIL3 utilizes a two‐dimensional hybrid parallel decomposition (Liu et al., 2014) replacing the one‐dimensional parallel decomposition in the meridional direction, increases the horizontal resolution from ~2.8° (128×60) to ~2° (180×80), and improves water vapor conservation through modification of the two‐step shape‐preserving advection scheme (TSPAS, Yu, 1994). With regard to physical processes, GAMIL3 incorporates a convective momentum transport scheme (Wu et al., 2007), adopts a simple stability‐based stratocumulus cloud fraction scheme based on estimated inversion strength (EIS; Guo & Zhou, 2014), and involves a simple parameterization of the second version of the Max Planck Institute Aerosol Climatology model (MACv2‐SP) for anthropogenic aerosol effects (Shi et al., 2019; Stevens et al., 2017) and an improved boundary layer scheme that includes entrainment at the top of the boundary layer, longwave radiative cooling at the top of stratocumulus clouds, and turbulent kinetic energy (TKE) (Sun et al., 2016). In addition, the external forcings recommended by CMIP6 were updated and their impacts on model stability, twentieth century global warming, and ENSO were evaluated by FGOALS‐g2 (Nie et al., 2019).

      CAS‐LSM is the land component of FGOALS‐g3 with the same horizontal resolution as the atmospheric component and is based on the Community Land Model Version 4.5 (CLM4.5). However, it includes unique improvements and additions to the land processes with respect to CLM4.5, such as groundwater lateral flow (Xie et al., 2012; Zeng, Xie, Yu, Liu, Wang, Jia, et al., 2016; Zeng, Xie, Yu, Liu, Wang, Zou, et al., 2016; Zeng et al., 2018), anthropogenic groundwater exploitation (Zeng, Xie, Yu, Liu, Wang, Zou, et al., 2016; Zeng et al., 2017; Zou et al., 2014, 2015), implementation of a new frozen soil parameterization including frost and thaw fronts (Gao et al., 2016, 2019), anthropogenic nitrogen discharge in rivers (S. Liu et al., 2019), and urban processes.

      LICOM3 is updated from LICOM2 (Lin et al., 2016; Liu et al., 2012). Its dynamical core with a latitude‐longitude grid structure is replaced by arbitrary orthogonal curvilinear coordinates (Madec & Imbard, 1996; Murray, 1996; Yu et al., 2018). Preserved shape advection (Xiao, 2006) and the implicit vertical viscosity (Yu et al., 2018) are used. The Laurent et al. (2002) tidal mixing model (Yu et al., 2017) is introduced into LICOM3. In addition, the eddy‐induced mixing of Redi (1982) and Gent and McWilliams (1990), and the buoyancy frequency (N2) related thickness diffusivity of Ferreira et al. (2005), were added to the model. The chlorophyll‐a‐dependent solar penetration of the Ohlmann (2003) scheme (Lin et al., 2007) and vertical mixing of Canuto et al. (2001, 2002) are used in LICOM3. A tripolar grid was chosen, with the North Pole split into two poles on‐land, which can enlarge the time steps in the Arctic polar region and remove the spatial filter for momentum velocities and tracers. A B‐grid was used for the horizontal distribution. The North Pole in the low‐resolution LICOM3 is divided into two North Poles on‐land at 65°N/65°E and 65°N/115°W. The low‐resolution LICOM3 has 360×218 horizontal grids. The vertical direction uses eta coordinates with 30 and 80 layers, but only the 30 layers were used for OMIP and CMIP6.

      The sea ice model is the Los Alamos sea ice model Version 4.0, using the same grid as the oceanic model. This is an energy conserving thermodynamic model, which solves the dynamic and thermodynamic equations for five ice thickness categories, with one snow and four ice layers. For the dynamic component, the elastic‐viscous‐plastic rheology (Hunke & Dukowicz, 1997), mechanical redistribution scheme (Lipscomb et al., 2007), and incremental remapping advection scheme (Lipscomb & Hunke, 2004) are used. For the thermodynamic component, the Delta‐Eddington radiative transfer scheme using inherent optical properties based on physical measurements (Briegleb & Light, 2007) is utilized.

      There are two couplers: CPL7 developed at the National Center for Atmospheric Research (NCAR) and C‐Coupler2 (Community Coupler Version 2) developed at Tsinghua University (Craig et al., 2012; Liu et al., 2018). Compared with CPL6 (Craig et al., 2005), CPL7 possesses improved memory and performance scaling that can support much higher‐resolution configurations. The computing performance of the coupled model using CPL7 was improved linearly by use of tens of thousands of CPUs. In addition, CPL7 has a more sophisticated computing resource control and a single executable, which allow the models to run flexibly and simplifies the machine requirements for the dispatcher. In addition to the Model Coupling Toolkit (MCT; Larson et al., 2005) that handles data transfer and interpolation for CPL7, C‐Coupler2 was employed as a new option for these two functionalities, which provides exactly the same (bitwise identical) simulation results as MCT. Moreover, the coupling capability of FGOALS‐g3 would be upgraded for future development with the new features of C‐Coupler2 (i.e., dynamic 3‐D coupling, flexible and automatic coupling generations, nonblocking data transfer, facilitation of increment coupling, and automatic remapping weight generation; Liu et al., 2018).