HTML
--> --> -->The Atmospheric Model Intercomparison Project(AMIP) simulation (Gates, 1992; Gates et al., 1999) is the first experiment designed in DECK. The AMIP experiment has been routinely carried out by modeling centers to evaluate their atmospheric models for its simplicity in methodology over the last three decades. The aim of the simulation is to analyze and evaluate the atmosphere and land in the climate system when they are constrained by observed sea surface temperatures (SSTs) and sea-ice concentrations. The systematic model errors can be identified by comparing the simulations to the observed atmosphere and land states in statistical ways. The simulation can also be useful for understanding climate variability and many aspects of historical climate changes for the climate science community.
The low-resolution version of the Chinese Academy of Sciences (CAS) Flexible Global Ocean-Atmosphere-Land System model, finite-volume version 3 (CAS FGOALS-f3-L) climate system model was developed at the State Key Laboratory of Numerical Modeling for Atmospheric Sciences and Geophysical Fluid Dynamics (LASG), Institute of Atmospheric Physics (IAP), CAS (Bao et al., 2019). The model completed the AMIP simulations in late 2018, and the model outputs were prepared for release after a series of postprocesses. To provide a description of the AMIP model outputs and the relevant essential model configurations and experimental methods for a variety of users, we document detailed descriptions of the AMIP simulation by CAS FGOALS-f3-L in this paper. Section 2 presents the model description and experimental design. Section 3 addresses the technical validation of the outputs from the CAS FGOALS-f3-L experiments. Section 4 provides usage notes.
2.1. Introduction to the model
CAS FGOALS-f3-L is composed of five components: version 2.2 of the Finite-volume Atmospheric Model (FAMIL) (Zhou et al., 2015; Bao et al., 2019; Li et al., 2019), which is the new generation atmospheric general circulation model of the Spectral Atmosphere Model of LASG (SAMIL) (Wu et al., 1996; Bao et al., 2010, Bao et al., 2013) (their main differences are shown in Table 1); version 3 of the LASG/IAP Climate system Ocean Model (LICOM3) (Liu et al., 2012); version 4.0 of the Community Land Model (CLM4) (Oleson et al., 2010); version 4 of the Los Alamos sea ice model (CICE4) (Hunke and Lipscomb, 2010); and version 7 of the coupled module from the National Center for Atmospheric Research (NCAR) (http://www.cesm.ucar.edu/models/cesm1.0/cpl7/), which is used to exchange the fluxes among these components.The atmospheric component, FAMIL, uses a finite-volume dynamical core (Lin, 2004) on a cubed-sphere grid (Putman and Lin, 2007), with six tiles across the globe. In FAMIL, each tile contains 96 grid cells (C96). Globally, the longitudes along the equator are divided into 384 grid cells, and the latitudes are divided into 192 grid cells, which is approximately equal to a 1° horizontal resolution. In the vertical direction, the model uses hybrid coordinates over 32 layers, and the model top is at 2.16 hPa. The main physical packages include a new moisture turbulence parameterization scheme for the boundary layer (Bretherton and Park, 2009), with shallow convection updated (Wang and Zhang, 2014). The Geophysical Fluid Dynamics Laboratory version of a single-moment six-category cloud microphysics scheme (Lin et al., 1983; Harris and Lin, 2014) is adopted to predict the bulk contents of water vapor, cloud water, cloud ice, rain, snow and graupel. For the cloud fraction diagnosis, the (Xu and Randall, 1996) scheme is used, which considers not only relative humidity but also the cloud mixing ratio, thus providing a more precise cloud fraction. A convection-resolving precipitation parameterization (\copyright 2017 FAMIL Development Team) is used where, in contrast to the conventional convective parameterization, convective and stratiform precipitation are calculated explicitly. The Rapid Radiative Transfer Model for GCMs (RRTMG) (Clough et al., 2005) was introduced into the model as the main radiation transfer, which utilizes the correlated k-distribution technique to efficiently calculate the irradiance and heating rate in 14 shortwave and 16 longwave spectral intervals. Finally, a gravity wave drag scheme is also used, based on (Palmer et al., 1986).
2
2.2. Experiments
Following the design of the DECK AMIP experiments (Eyring et al., 2016), we conducted three simulations, as summarized in Table 2. In these experiments, the external forcings are prescribed as their monthly mean observation values, as recommended by the CMIP6 projects: the historical global mean greenhouse gas concentrations from (Meinshausen et al., 2017); solar forcing from (Matthes et al., 2017); historical ozone concentrations from http://blogs.reading.ac.uk/ccmi/forcing-databases-in-support-of-cmip6/; and AMIP SST and Sea Ice Datasets from the program for Climate Model Diagnosis & Intercomparison (PCMDI) at https://esgf-node.llnl.gov/projects/esgf-llnl/. The aerosol mass concentrations are also prescribed and taken from the NCAR Community Atmosphere Model with Chemistry (CAM-Chem; Lamarque et al., 2012), there are five aerosol species including sulfates, sea salts, black carbon, organic carbon, and dust. The land use datasets are prescribed as their mean climate values (Hurtt et al., 2011).As shown in Table 2, the experiment_id and variant_label are presented to identify each experiment and the corresponding outputs (Table 3). The time-lag method is used to realize the three perturbations that were identified by the variant_label: r1i1p1f1, r2i1p1f1, and r3i1p1f1. The characteristics in r1i1p1f1 denote the realization_index, initialization_index, physics_index, and forcing_index. The three ensemble simulations share the same model physics and forcing but differ due to their different integration start dates. The first experiment (r1i1p1f1) integrates from 1 January 1970. The first nine years are considered to be the spin-up period, and the model outputs from 1979 to 2014 are provided for public users. The second experiment (r2i1p1f1) is the same as r1i1p1f1, except that the integration start date is 1 January 1971. Similarly, the start date is 1 January 1972 in the third experiment (r3i1p1f1). All simulations are forced by the same varying external forcing during the observed time listed in the last paragraph.
Figure1. Climatological (1979-2014) annual mean precipitation (units: mm d-1) from (a) GPCP and (b) the mean of amip r1i1p1f1, r2i1p1f1, and r3i1p1f1.The simulation of the Madden-Julian Oscillation (MJO) is of interest in current climate models and has remained a great challenge in recent years (Jiang et al., 2015). Here, we present the model skill in capturing the MJO based on daily precipitation and 850 hPa winds. The observed daily precipitation from GPCP (Huffman et al., 2001) and the wind field from ERA-Interim (Dee et al., 2011) are used as reference observations. Using a 20-100-day band-filtered component, we analyzed the zonal propagation of precipitation (colors) and 850-hPa zonal winds (contours) against precipitation in an Indian Ocean reference region (10°S-5°N, 75°-100°E) for boreal winter (Fig. 2). Here, winter is defined from November to April of the following year, following (Waliser et al., 2009). Compared with the observations, the dominant feature of MJO eastward propagations (from the Indian Ocean via the western Pacific to the International Date Line) can be simulated well in both precipitation and 850 hPa winds in the AMIP simulation. The quadrature relationship between precipitation and the 850 hPa zonal winds (U850) is reproduced well over the Indian Ocean and western Pacific Ocean in the simulation. Meanwhile, the phase speed is nearly 4-5 m s-1, and the lag of the wind anomaly behind precipitation is approximately 5-7 days in the simulation, which is also similar to the result observed in (Waliser et al., 2009). Compared with the previous version of FAMIL (Yang et al., 2012), the MJO simulation is substantially improved. Small weaknesses are also identified. The propagation of precipitation is increased by two to three days compared to the observation reaching the date line. The rainfall amplitude associated with the MJO is also slightly weaker than in the observation.
Figure2. November-April lag-longitude diagram of the 10°S-10°N intraseasonal precipitation anomalies (colors) and intraseasonal 850-hPa zonal wind anomalies (contours) correlated with intraseasonal precipitation over the Indian Ocean reference region (10°S-5°N, 75°-100°E) for (a) observations and (b) the mean of amip r1i1p1f1, r2i1p1f1, and r3i1p1f1.Tropical cyclones (TCs), as one of the most drastic phenomena in the world, have considerable impacts on human life. TC forecasting is still a challenge in that most models are unable to predict the tracks very well (Xiang et al., 2015). We evaluate the simulations of TC tracks in AMIP r1i1p1f1 based on the six-hourly datasets in Fig. 3. The observed TC tracks (Fig. 3a) are derived from the International Best Track Archive for Climate Stewardship (IBTrACS) (version v03r09) dataset (Knapp et al., 2010), which shows that TCs are active in subtropical oceans in both hemispheres, except for the southeastern Pacific Ocean and the southern Atlantic Ocean. TCs were classified into seven categories according to the Saffir-Simpson (SS) scale (Simpson and Saffir, 1974), as shown in different colors in Fig. 3. Based on the six-hourly dataset in AMIP r1i1p1f1, the model could successfully capture the global pattern of the tropical storm (TS) tracks (green lines) (Fig. 3b), except that the model underestimates the TS tracks in the eastern Pacific and northern Atlantic Ocean while producing unrealistic tracks over the southern Atlantic. Category 4 and 5 TCs are also underestimated in the model. These results suggest that the model can capture TC tracks, but the intensity is slightly weaker, and the six-hourly datasets are quite reliable for conducting TC research.
Figure3. TC tracks (lines) and intensities (colors) (a) from a 36-year segment (1979-2014) of IBTrACS data and (b) from the simulation of AMIP r1i1p1f1, which are detected by using an objective feature-tracking approach at the C96 resolution (approximately 100 km) from 1979-2014. Only those TCs with a lifetime exceeding three days are shown. The TCs in both the simulation of AMIP r1i1p1f1 and the observation of IBTrACS are grouped into seven categories in accordance with the modified Saffir-Simpson scale, but only TCs stronger than TSs are shown.Realistic reproduction of historical extreme precipitation has been challenging for both reanalysis and GCM simulations (He et al., 2019). We evaluate the simulations of extreme precipitation over the tropics (20°S-20°N) in the AMIP r1i1p1f1 daily outputs. Tropical Rainfall Measuring Mission (TRMM)-3B42 data are used as reference observations (Huffman et al., 2007) and were interpolated at the same resolution as AMIP r1i1p1f1 by the nearest-neighbor interpolation method (Accadia et al., 2003). The frequency of precipitation was plotted against the daily precipitation rate at a 1 mm d-1 interval. For the extreme precipitation, the frequency-intensity distribution (Fig. 4) in CAS FGOALS-f3-L is extended up to 350 mm d-1, which shows that similar characteristics manifested in the TRMM data. The frequency-intensity distribution in the model is also comparable under 50 mm d-1 with the TRMM data, while it is slightly overestimated above 50 mm d-1. These results suggest that the model can simulate enough extreme precipitation over tropical regions, but the frequency of extreme precipitation is slightly overestimated.
Figure4. Annual rainfall frequency-intensity distribution in TRMM-3B42 (black line) and CAS FGOALS-f3-L AMIP r1i1p1f1 (blue line) from 1998-2010. The domain is the tropical region (20°S-20°N).The format of datasets is the version 4 of Network Common Data Form (NetCDF), which can be easily read and written by professional common software such as Climate Data Operators (https://www.unidata.ucar.edu/software/netcdf/workshops/2012/third_party/CDO.html), NetCDF Operator (http://nco.sourceforge.net), NCAR Command Language (http://www.ncl.ucar.edu), and Python (https://www.python.org).
3
Data availability statement
The data that support the findings of this study are available from https://esgf-node.llnl.gov/projects/cmip6/.3
Disclosure statement
No potential conflict of interest was reported by the authors.Open Access This article is distributed under the terms of the Creative Commons Attribution License which permits any use, distribution, and reproduction in any medium, provided the original author(s) and the source are credited.
