The Nonhydrostatic ICosahedral Atmospheric Model (NICAM), a global model with an icosahedral grid system, has been under development for nearly two decades. This paper describes NICAM16-S, the latest stable version of NICAM (NICAM.16), modified for the Coupled Model Intercomparison Project Phase 6, High Resolution Model Intercomparison Project (HighResMIP). Major updates of NICAM.12, a previous version used for climate simulations, included updates of the cloud microphysics scheme and land surface model, introduction of natural and anthropogenic aerosols and a subgrid-scale orographic gravity wave drag scheme, and improvement of the coupling between the cloud microphysics and the radiation schemes. External forcings were updated to follow the protocol of the HighResMIP. A series of short-term sensitivity experiments were performed to determine and understand the impacts of these various model updates on the simulated mean states. The NICAM16-S simulations demonstrated improvements in the ice water content, high cloud amount, surface air temperature over the Arctic region, location and strength of zonal mean subtropical jet, and shortwave radiation over Africa and South Asia. Some long-standing biases, such as the double intertropical convergence zone and smaller low cloud amount, still exist or are even worse in some cases, suggesting further necessity for understanding their mechanisms, upgrading schemes and parameter settings, and enhancing horizontal and vertical resolutions.
Moist processes play a crucial role in the formation of the earth's climate. Moist convection redistributes mass, energy, and momentum of the atmosphere to form large-scale circulation. Clouds are coupled with large-scale circulation through latent and radiative heating, which can affect climate sensitivity. The accurate treatment of such interactions between clouds and circulation requires high-resolution global cloud-resolving models (Bony et al., 2015; Satoh et al., 2019).
This, as well as an increasing demand from society to project extreme weather, such as tropical cyclones, motivated us to perform climate simulations of the present day and future using the Nonhydrostatic ICosahedral Atmospheric Model (NICAM) (Tomita and Satoh, 2004; Satoh et al., 2008, 2014) with a 14 km mesh. Kodama et al. (2015) and Satoh et al. (2015) provided brief descriptions of the model (hereafter referred to as NICAM.12) and experimental design of the climate simulations. This unique dataset of the high-resolution climate simulation, whose overall performance was reported in Kodama et al. (2015), has been used in many studies with a focus on tropical cyclones (Satoh et al., 2015, 2018; Yamada et al., 2017, 2019; Matsuoka et al., 2018; Sugi et al., 2020), extratropical cyclones (Kodama et al., 2019; McCoy et al., 2019; Satoh et al., 2018), intraseasonal oscillations, such as the Madden–Julian Oscillation (MJO) (Kikuchi et al., 2017; Nakano and Kikuchi, 2019), tropical synoptic-scale waves (Fukutomi et al., 2016), cloud radiative feedback (Chen et al., 2016; Satoh et al., 2018; Noda et al., 2019), and regional to global precipitation (Kilpatrick et al., 2017; Satoh et al., 2018; Na et al., 2020; Takahashi et al., 2020).
Also, some significant climate biases have been identified in the simulation (Kodama et al., 2015), and great effort has been devoted to improving the model for better performance of the simulated climate in a physically consistent manner. Major updates from NICAM.12 to NICAM.16, a stable version of NICAM released in 2017, included the cloud microphysics scheme based on a comparison with the Tropical Rainfall Measuring Mission (TRMM) satellite observations (Roh and Satoh, 2014; Roh et al., 2017), introduction of a wetland scheme in the land surface model (Nitta et al., 2017), implementation of the coupling between cloud microphysics and radiation that considers the nonsphericity of ice particles (Seiki et al., 2014), and introduction of a subgrid-scale orographic gravity wave drag scheme. In addition, some parameters related to the processes of surface albedo and sea ice have been revised. These model updates generally reduce the biases of the simulated mean states, as discussed below. NICAM.16 has been further modified to support the external forcings of natural and anthropogenic aerosols and the solar cycle defined in the Coupled Model Intercomparison Project Phase 6 (CMIP6) High Resolution Model Intercomparison Project (HighResMIP) protocol (Haarsma et al., 2016). This special version of NICAM.16 for HighResMIP is labeled NICAM16-S, where “-S” represents the use of a single-moment cloud microphysics scheme. A double-moment cloud microphysics scheme is also available in NICAM.16 (Seiki and Nakajima, 2014; Seiki et al., 2014, 2015b; Satoh et al., 2018). However, the double-moment scheme was not used for the HighResMIP simulations and hence is not described in this paper. NICAM.16 is not coupled with an ocean model, and the DECK (Diagnostic, Evaluation and Characterization of Klima) and CMIP historical simulations (Eyring et al., 2016) are not presented in this study. Note that a coupled ocean–atmosphere model, NICAM-COCO (Miyakawa et al., 2017), is being developed.
This section has provided a summary of NICAM16-S with a focus on its differences from NICAM.12. Section 2 of this paper presents the experimental design of the HighResMIP simulations and a series of sensitivity experiments by NICAM16-S. Section 3 explains the detailed model updates of NICAM16-S from NICAM.12 and their impacts on the simulated mean states. Section 4 briefly describes the resolution dependency of NICAM16-S. Section 5 reports on the computational aspects of the simulation. Section 6 provides a brief summary of the comparison results.
Three sets of model configurations were prepared for the HighResMIP
simulations, and initial and boundary conditions were made for each.
NICAM16-S models with specific horizontal resolutions were formally labeled
NICAM16-7S (56 km mesh), NICAM16-8S (28 km mesh), and NICAM16-9S (14 km
mesh) in CMIP6 as source IDs. The horizontal mesh size is evaluated as a
square root of the mean area of each grid cell
(Satoh et al., 2014). The number
The number of vertical levels is 38, with a model top height of around 40 km, equivalent to the previous climate simulations (Kodama et al., 2015). The interval between each vertical layer increases from 160 m to 2 km as the altitude increases from the ground to 25 km (see the “K38” setting in Fig. 1 of Ohno et al., 2019). Even at such a low vertical resolution, atmospheric phenomena of interest, including tropical cyclones, the MJO, and the diurnal precipitation cycle, may be simulated practically and accurately, as we have confirmed in the previous study using a 14 km horizontal mesh (Kodama et al., 2015). As a caveat, a coarse vertical resolution in the upper atmosphere leads to an overestimation of the cirrus cloud amount (Seiki et al., 2015b; Ohno et al., 2019) and may cause a different response of high cloud to warmer climate (Ohno et al., 2019). Also, it has been suggested that the vertical resolution should be increased when the horizontal resolution is increased in terms of atmospheric gravity waves (Lindzen and Fox-Rabinovitz, 1989; Polichtchouk et al., 2019). Such coarse vertical resolution in this study could change vertical propagation of gravity waves and zonal wind in the stratosphere (Watanabe et al., 2015; Skamarock et al., 2019).
The time step of the dynamics (
Table 1 shows the integration periods for the HighResMIP simulations. For the Tier 1 simulations, which start from 1 January 1950, the initial condition of the atmosphere was taken from the ERA-20C reanalysis (Poli et al., 2016). The NICAM16-7S and NICAM-8S simulations strictly followed the HighResMIP protocol and continued until 31 December 2014. The HighResMIP Tier 3 simulations using NICAM16-7S and NICAM16-8S started on 1 January 2015 as a continuation of the Tier 1 simulations and ended on 31 December 2050. The high computational cost hindered us from running NICAM16-9S for 100 years and thus a time slice approach was adopted instead. Specifically, climate simulations were performed in the following timeframes: 1950–1960, 2000–2010, and 2040–2050. For the simulations starting from 1 January 1950 or 1 January 2000, the initial land conditions prescribed for NICAM16-7S, NICAM16-8S, and NICAM16-9S were taken from the monthly mean climatology of the simulation by NICAM with a mesh size of 220 km (glevel 5) under present-day conditions. It was performed for 10 years, and the last 5 years of data were used to obtain the monthly mean land climatology. This approach is the same as that used in previous climate simulations (Kodama et al., 2015). This could partly reduce the initial shock of the land surface model, even though it may cost more than several years for some land variables, such as soil moisture, to fully settle down (not shown). The initial land condition for the future time slice run with NICAM16-9S was obtained by interpolating the output of the NICAM16-8S simulation.
List of HighResMIP simulations.
List of sensitivity experiments.
In addition to the formal HighResMIP simulations, we performed a series of short-term sensitivity experiments to evaluate the impacts of the model changes on the simulated climatology, as listed in Table 2. Here, the model configuration of the REFFIX run is equivalent to that used in the formal HighResMIP simulations. As noted in Sect. 3.7, we often prefer to use a slab ocean model with nudging toward the boundary sea surface temperature (SST) rather than the fixed SST condition requested by the HighResMIP protocol because of better performance in the simulated precipitation pattern (Kodama et al., 2015), particularly with a horizontal mesh size of 14 km (Sect. 3.7). Therefore, both the fixed SST and slab ocean configurations (REFFIX and REFSLB runs, respectively) were tested in the sensitivity experiments, and we used the REFFIX run with 56 km mesh and the REFSLB run with 14 km mesh as the reference (REF) runs for the other sensitivity experiments. Impacts of the model updates described in Sect. 3 on the simulated climate states were individually tested by switching off each update. Additionally, the DDT2M, DDT1M, RDT20M, and RDT10M runs were performed to check the sensitivity of the simulated climate to the time step of the model. These sensitivity experiments were started on 1 June 2004 and integrated for 1 year. An exception was the NOLND and the REF runs, which were performed for 4 years to make the land surface state settle down. The initial date was chosen to ensure consistency with previous NICAM studies (e.g., Kodama et al., 2012; Seiki et al., 2015a; Noda et al., 2016). An integration period of 1 year, or even less, is sufficient to evaluate the basic state of the atmosphere, such as cloud amount, precipitation, radiation, and temperature (e.g., Phillips et al., 2004; Noda et al., 2010; Kodama et al., 2012; Williams et al., 2013; Miyakawa et al., 2018; Miyakawa and Miura, 2019; Stevens et al., 2019; Hohenegger et al., 2020), and tropical variability, including diurnal cycle, tropical cyclones, and MJO (e.g., Sato et al., 2009; Kinter et al., 2013; Stevens et al., 2019; Matsugishi et al., 2020). The interannual variability of the HighResMIP simulation by NICAM16-7S (Table 1) over 101 years was diagnosed to distinguish the impacts of the model changes from internal variability in a rough manner. The simulation data were regridded to the same grid of observations (Table 3) unless it is explicitly specified.
List of observational datasets.
External forcings and boundary conditions of the simulations followed the HighResMIP protocol (Haarsma et al., 2016). Historical and SSP5-8.5-scenario settings (O'Neill et al., 2016) were used in the Tier 1 and 3 simulations, respectively.
Daily quarter-degree SST and sea ice mass (ICE) prescribed for the model were obtained from HadISST 2.2.0.0 (Kennedy et al., 2017). The SST dataset was extended from 2016 to 2050 using the trend obtained from a CMIP5 model ensemble mean following the RCP8.5 scenario and historical variability from 1980 to 2015 (Kennedy et al., 2019). Because HadISST provides historical sea ice concentration (SIC), ICE was diagnosed from SIC for NICAM (see Sect. 3.7). The future SIC was estimated from the future SST data and the observed relationship between SST and SIC (HighResMIP, 2020; Kennedy et al., 2019). Both the SST and ICE were fixed to the boundary conditions in the HighResMIP Tier 1 and 3 simulations.
Figures 1 and 2 display the decadal mean SST
and ICE, respectively, prescribed for the model;
Fig. 3 exhibits their global mean variability. Greater warming over the
maritime continent, the Indian Ocean, and the edge of the polar regions are
found in the 2000s compared with the 1950s, whereas cooling is noticed in
the North Pacific and the North Atlantic. The SST in the 2040s has larger
values almost everywhere compared with that in the 2000s, especially in the
midlatitudes, the equatorial eastern Pacific, the tropical Atlantic Ocean,
and the edge of the Arctic regions. Similar tendencies are observed for the
distribution of the SST trend (not shown). The global mean SST is
17.9
Decadal mean horizontal distribution of sea surface temperature
(SST) prescribed in the model averaged for the 1950s
Same as Fig. 1 but for sea ice mass (ICE)
in kg m
The global annual mean of greenhouse gas (GHG) concentrations
(Meinshausen
and Vogel, 2016; Meinshausen et al., 2017; Meinshausen and Nicholls, 2018)
was prescribed in the model. Specifically, CO
Global mean SST (
Mass and number concentrations of natural aerosols prescribed in the model
were obtained from a low-resolution NICAM simulation with an online aerosol
module based on the Spectral Radiation-Transport Model for Aerosol Species
(SPRINTARS)
(Takemura
et al., 2000, 2002, 2005, 2009; Goto et al., 2008, 2011). The simulated
climatology of aerosols in NICAM was previously validated
(Goto et al., 2018; Suzuki et al., 2008). For
HighResMIP, NICAM with a mesh size of 220 km was performed for 100 years
using natural aerosol emissions with the anthropogenic aerosol module
MACv2-SP (Fiedler et al.,
2017; Stevens et al., 2017; see Sect. 3.4) under a perpetual 2012
condition, and the data for the last 90 years were averaged to obtain a
monthly mean climatology of aerosol mass and number concentrations of cloud
condensation nuclei (CCN) from a natural origin. A lower-bound limiter of
50 cm An error was found in the natural aerosol forcing prescribed
in the model, as recently reported on the ES-DOC Errata website (
Similar to the implementation of MIROC6 (Tatebe et al., 2019), historical monthly mean solar forcings (Matthes et al., 2017b, a) were prescribed as total solar irradiance and solar irradiance spectra in the radiation scheme mstrnX (Sekiguchi and Nakajima, 2008). In terms of land surface processes, a monthly climatology (2004–2013) of leaf area index was obtained from the Moderate Resolution Imaging Spectroradiometer (MODIS) product (MCD15A2.005, Shabanov et al., 2005; Yang et al., 2006).
As briefly noted in Satoh et al. (2014), a
spatial filter was applied to smooth the model topography to avoid numerical
instability. Specifically, a hyper-diffusion was repeatedly applied to the
GTOPO30 (
The dynamical core and numerical filters in NICAM16-S are the same as those in NICAM.12. NICAM adopts a fully compressible nonhydrostatic system as the governing equations of the dynamics (Tomita and Satoh, 2004; Satoh et al., 2008). The horizontal discretization is an icosahedral grid system modified with spring dynamics for homogeneity on the sphere (Tomita et al., 2002). Divergence damping and second-order Laplacian horizontal diffusion are used to stabilize the integration (Satoh et al., 2008). Additionally, first-order Laplacian horizontal diffusion is applied above an altitude of 20 km to avoid spurious wave reflection at the model top.
Table 4 gives a summary of the physics schemes used
in NICAM16-S and NICAM.12. A single-moment bulk cloud microphysics scheme,
NICAM Single-Moment Water 6 (NSW6), that solves mass concentrations for six
water categories – vapor, cloud water, cloud ice, rain, snow, and graupel
(Tomita, 2008; Roh and
Satoh, 2014; Roh et al., 2017) – is used instead of a combination of
convection and large-scale condensation schemes. While most climate models
use convection and large-scale condensation schemes, even for a mesh size
around 14 km, we used the cloud microphysics scheme to represent
interactions between clouds and circulation in an explicit way. This not
only lowers the cost of model development but also reduces the uncertainty
in the results arising from highly arbitrary tuning. This approach has also
been tested by other researchers besides the NICAM group
(Maher et al., 2018; Hohenegger et al.,
2020). Global mean precipitation is constrained by radiative cooling in
large-scale clear-sky regions, which can be captured by a model with
relatively coarse resolution without the convection and large-scale
condensation schemes. The simulated climatology of the precipitation
pattern, even with the lowest resolution setting (NICAM16-7S), is comparable
with the observed patterns, as shown below, although our choice leads to
patchy precipitation behavior and dry/wet bias in the middle/lower
troposphere in the simulation (Miyakawa et al., 2018).
Similar precipitation behavior was also reported in a climate model study
with a mesh size of around O(10
Physics schemes in NICAM16-S and NICAM.12.
A modified version of the Mellor–Yamada level 2 scheme (Nakanishi and Niino, 2006; Noda et al., 2010) is used to simulate the planetary boundary layer. The radiation scheme, mstrnX (Sekiguchi and Nakajima, 2008), is a broadband model with 29 radiation bands as used here. The land surface model, Minimal Advanced Treatments of Surface Interaction and RunOff (MATSIRO) (Takata et al., 2003) solves land states such as soil temperature, soil moisture, and land surface fluxes. Ocean surface fluxes are calculated following Louis (1979) with a modified roughness length for strong surface wind conditions (Fairall et al., 2003; Moon et al., 2007). The conventional orographic gravity wave drag scheme (McFarlane, 1987) is used to introduce the effect of vertically propagating subgrid-scale orographic gravity waves on the momentum tendency of the atmosphere. Though we did not fine-tune the model due to heavy computational cost, we crudely turned parameters of sea ice thickness with NICAM16-7S (Sect. 3.7) and the gravity wave drag scheme with NICAM16-9S (Sect. 3.8). We did not return the model at each resolution under the principles of the HighResMIP.
As we discuss in Sect. 2.2 and Table 2, a series of short-term sensitivity experiments were performed to monitor the impacts of several model updates on the simulated climatology. Figure 5 (right part of each panel) summarizes the impacts of the model changes on the global mean climate. All the significant impacts of the model changes shown here could be qualitatively reproduced even when the analysis period was limited to the last 6 months (not shown). The REFFIX and REFSLB runs with each horizontal mesh and the observations are shown on the left in each panel in Fig. 5. We will discuss these impacts along with the details of the model updates later in this section.
Global annual means of surface air temperature
Recently, Roh and Satoh (2014) and Roh et al. (2017) significantly revised the NSW6 scheme based on comparisons with TRMM observations. We used the revised version of the NSW6 scheme in NICAM16-S and found improvements in the simulated climatology, as shown below.
The NSW6 scheme originated with Lin et al. (1983) and Rutledge and Hobbs (1984). Tomita (2008) modified their cloud microphysics scheme to ensure consistency with the thermodynamics used in NICAM. Tomita (2008) also simplified it to reduce the calculation cost for high-resolution global simulations. NSW6 was evaluated by comparing the simulated optical properties with satellite observations (Satoh et al., 2010; Kodama et al., 2012; Hashino et al., 2013, 2016; Roh and Satoh, 2014, 2018; Roh et al., 2017) using satellite simulators, specifically, the CFMIP Observational Simulator Package (Haynes et al., 2007; Chepfer et al., 2008; Bodas-Salcedo et al., 2008, 2011) and Joint Simulator (Matsui et al., 2009; Masunaga et al., 2010; Hashino et al., 2013). It was revised in each stage of the version management of NICAM (Kodama et al., 2012; Roh and Satoh, 2014; Roh et al., 2017). The revision of the NSW6 scheme by Roh and Satoh (2014) and Roh et al. (2017) represents a significant change between NICAM.12 and NICAM16-S.
Table 5 summarizes key changes in the NSW6 scheme by Roh and Satoh (2014) and Roh et al. (2017). In short, the revision aimed to enhance organizations of tropical convective cloud systems by assuming lighter precipitation of graupel and snow and by moderating the development of cloud ice. Finally, they successfully reproduced the vertical structures of shallow, congestus, and deep convective clouds over the tropics compared to TRMM and CloudSat satellite observations. They also used microwave satellite observations to evaluate the simulated results (Roh and Satoh, 2018). Therefore, improvements in tropical cloud systems with the revised scheme are robust in terms of optical signals (see the original papers for more details). Note that the separation of convective and stratiform systems in Roh and Satoh (2014) was omitted in this study because of its small impact (not shown) despite the high computational cost.
Summary of the key changes in the NSW6 scheme by Roh and Satoh
(2014) and Roh et al. (2017).
The sensitivity experiments with and without the update of the cloud microphysics scheme (the REF and NOCLD runs, respectively) were compared. The most noticeable impact was an increase in ice water content (IWC) by more than 2-fold (Fig. 5e), and this mostly accounts for the snow category in the cloud microphysics scheme. Figure 6 shows the meridional-height cross section of the observed and simulated zonal mean IWC and its breakdown into the categories of cloud ice, snow, and graupel in the NOCLD and REF runs. The simulated IWC is largely underestimated using the model without the updated cloud microphysics scheme (the NOCLD run), as also shown in Seiki et al. (2015a), and it becomes comparable with CloudSat observations with the update (the REF run). A noticeable increase in the snow category was seen in the tropical upper troposphere and midlatitude storm-track region. Cloud ice is also increased in the upper troposphere. Graupel increased in the tropical middle troposphere but decreased in the storm-track region. As a result, global mean column-integrated cloud ice and snow increased by 24 % and 399 %, respectively, and that of graupel decreased by 8.4 %. The net increase in IWC is consistent with the decelerated development of IWC by the modified mass and diameter relationship of snow, which reduces the snow density (Table 5, d) by the diminished efficiency of accretion of cloud ice by snow (Table 5, g) and by ignoring the accretion of snow and cloud ice by graupel (Table 5, f). The increase/decrease in graupel by the model update may be linked to the differences in cloud processes (convection versus synoptic systems).
Annual mean of the zonal mean ice water content (10
Despite the drastic increase in mass concentrations of snow and cloud ice,
the amount of high clouds, particularly, optically thin clouds, is reduced
by the update (Fig. 5g). Consistently, global
mean outgoing longwave radiation (OLR) is increased by about 4 W m
TOA brightness temperature near 11
In NICAM16-S, the cloud microphysics scheme is fully coupled with the
radiation scheme mstrnX (Sekiguchi and Nakajima,
2008). The effective radius of hydrometeors is calculated with the same
assumption of the particle size distribution function as the cloud
microphysics scheme, including indirect effects, and then passed to the
mstrnX scheme. In contrast, fixed effective radii of 8
The mstrnX scheme requires a database of single-scattering properties of
hydrometeors (RADPARA), including parameters such as the volume extinction
coefficient, absorption coefficient, asymmetry factor, and second moment of
phase function (Nakajima et al., 2000). In NICAM16-S we
used the RADPARA database revised by Seiki et al. (2014).
The RADPARA database of liquid hydrometeors was precalculated according to
the Mie theory. The nonspherical RADPARA database developed by
Fu (1996) and Fu et al. (1998) was
applied to solid hydrometeors. The RADPARA database was then compiled as a
lookup table of the effective radii from 1
Because of nonsphericity, the effective radius of ice particles has a
controversial definition, whereas the effective radius is well defined in
the case of spherical particles. According to Fu (1996),
the effective radius,
The impacts of the coupling procedure and the nonspherical scattering were examined by comparing the REF run with the NONSI run using the fixed effective radii and the spherical RADPARA database. The impacts of nonspherical scattering alone could be seen from the comparison between the REF and NONS runs using the spherical RADPARA database. Figure 8 shows the zonal mean values of the OLR and the outgoing shortwave radiation (OSR) at the TOA from the sensitivity experiments. Given the substantial increase in cloud ice and snow from the revised NSW6 scheme (cf. Sect. 3.2), ice optical thickness increases proportionately to the increases in the ice water path with fixed effective radii. As a result, both the longwave and shortwave radiation budgets are strongly biased in the NONSI run. A major portion of the biases in OLR is drastically offset in the NONS run by assuming larger effective radii in the coupling procedure. Thus, the coupling procedure automatically prevents artificial biases originating from inconsistent parameter settings between the cloud microphysics and radiative transfer with the model update. The use of the nonspherical RADPARA database slightly increases OSR over the tropics to the midlatitude because the assumed asymmetry factor for nonspherical particles is smaller than that for spherical particles (cf. Seiki et al., 2014).
Annual mean of the zonal mean OLR
Finally, NICAM16-S still shows strong negative biases in OLR over the tropical to subtropical regions and OSR over the subtropical high-pressure belt at the TOA compared with the Clouds and Earth's Radiant Energy Systems (CERES) product (Fig. 8, black versus red curves), and these biases are qualitatively similar to those simulated in NICAM.12 (Kodama et al., 2015). The former OLR bias can be solved by increasing vertical resolution to 400 m near the tropopause with 74 vertical layers (Seiki et al., 2015b). The latter OSR bias mainly stems from the underestimation of low-level clouds, and the current updates in the cloud microphysics scheme do not improve the results for warm clouds. The coupling procedure strengthens the negative biases in OSR at TOA in the midlatitudes (Fig. 8, green versus blue curves) because, in NICAM.12, high clouds associated with extratropical cyclones are artificially brightened and, therefore, conceal the biases due to low-level clouds (cf. Fig. 4 in Kodama et al., 2012). Unlike NICAM.12, a strong negative bias of OSR at the TOA is also prominent over the Arctic region, and this seems to relate to the update (reduction) of the surface albedo introduced in Sect. 3.6.
In NICAM.12, the direct radiative effect of aerosols is not considered in
the radiation scheme, and the number concentration of CCN is set to a
constant value of 50 cm
The sensitivity experiments with and without the update of the aerosol
treatment (the REF and NOAER runs) are compared in Fig. 9. The aerosol
update reduces the net downward shortwave radiation at the surface over most
of the continents, particularly over Africa and South Asia (Fig. 9b and
d), leading to a reduction of the excess insolation there (not shown). The
reduced insolation at the surface is partly canceled out by a reduction in
the upward longwave radiation over Africa (Fig. 9a and c) in association
with a decrease in surface air temperature (not shown). Meanwhile, an
enhancement of the surface net downward shortwave radiation is dominant over
the ocean, particularly in the NICAM16-9S (14 km mesh) run (Fig. 9b and
d), in association with a thinning of cloud optical depth and a decrease in
cloud amount (not shown). This links to a decrease in the liquid water path
with the aerosol update (Fig. 5f), which was also
found in an online aerosol experiment with the 14 km mesh NICAM
(Sato et al., 2018). As a result of these compensations, the
global mean net surface radiation change arising from aerosol forcing is
around
Impact of the aerosol treatment update (the REF run minus
the NOAER run). Two-dimensional distribution, zonal, and global means of the net
longwave
The land surface model MATSIRO (Takata et al., 2003) is used in NICAM. Recently, a wetland scheme was implemented in MATSIRO of NICAM16-S to represent the storage of snowmelt while considering the subgrid-scale terrain complexity (Nitta et al., 2017). The wetland scheme reduces the summertime warm and dry bias over much of Western Eurasia and North America through delayed snowmelt runoff in MIROC5 (Nitta et al., 2017). In addition, the effect of decreased surface albedo associated with the accumulation of water on land ice was implemented in NICAM16-S.
Figure 10 shows the impact of the land surface model update on soil moisture during boreal summers. The soil moisture is increased over most of the Eurasian and North American continents, as expected from Nitta et al. (2017), particularly in Siberia and around the Great Lakes. Though it is expected from Nitta et al. (2017) that increased soil moisture leads to an increase in precipitation and a decrease in surface air temperature, the actual impacts on precipitation and surface air temperature are still unclear (not shown). It is difficult to show robust reduction of the biases at this stage, and longer integration is needed to assess these impacts appropriately.
Impact of the land surface model update (the REF run minus
the NOLND run) on the simulated soil moisture at the uppermost land surface
model level. Simulations using NICAM16-7S were performed for 4 years, and
June–July–August data for the last 3 years are averaged. The data are
gridded at 2.5
Surface albedo values were updated based on observations. In NICAM.12, they were tuned to reduce the TOA radiation imbalance, and this caused a higher bias of surface albedo over the Arctic compared with that seen in satellite observations (Hashino et al., 2016). Table 6 shows the surface albedo values used in NICAM16-S and NICAM.12. The albedo values of the sea ice for the visible and near-infrared wavelengths and of fresh snow over land for the visible wavelengths were set to be smaller in NICAM16-S than those in NICAM.12. In addition to the updates in Table 6, an artificial elevation of the ocean surface albedo for the direct visible wavelength by a factor of 1.35 for the radiation scheme in NICAM.12 was discarded in NICAM16-S.
Surface albedo in NICAM16-S and NICAM.12.
VIS, NIR, and IR stand for visible, near-infrared, and infrared bands, respectively.
The sensitivity experiments with and without the albedo update (the REF and NOALB runs, respectively) show that the use of the new surface albedo values tends to reduce the surface air temperature bias over land ice compared with the previous bias (Fig. 11a versus Fig. 11c and green versus red curves in Fig. 11d). Specifically, the cold bias in Greenland, the Himalayas, and the Antarctic is reduced. This is consistent with the reduced surface albedo for the visible wavelengths and the resulting decreased net upward shortwave radiation at the surface (Fig. 5k). Global mean net downward longwave radiation at the surface is increased (Fig. 5j). The increase is attributed primarily to a decrease in upward longwave radiation over the ocean (not shown), consistent with the increased surface albedo for the infrared wavelengths over the ocean. In terms of the TOA radiation budget, OSR worsens by a few watts per square meter (Fig. 5d), which arises from the polar regions (Fig. 12b; green versus red curves).
Bias in the simulated surface air temperature (K) by NICAM16-7S
against JRA-55 reanalysis averaged for June 2004–May 2005. The NOALB
Same as Fig. 8 but for the NOALB (green), NOSIC (blue), and REF (red) runs by NICAM16-7S.
A mixed-layer slab ocean model similar to that in
McFarlane et al. (1992) was implemented in
NICAM. The model predicts SST, ICE, snow over sea ice, and snow temperature
by solving the heat balance between the ocean, sea ice, snow, and
atmosphere. The depth of the slab ocean model is set to 15 m globally,
considering the better performance of the simulated precipitation pattern
(Kodama et al., 2015) and MJO
(Grabowski, 2006). A simple nudging
technique is used to force the predicted SST and ICE toward their reference
states with a relaxation time of
In the slab ocean model implemented in NICAM.12 and NICAM16-S, SIC is
diagnosed from ICE as
In many cases, including the HighResMIP protocol, only the SST and SIC data
are provided to run the model, and, therefore, the ICE data prescribed for
the model should be diagnosed from SIC data. In NICAM.12 and NICAM16-S, ICE
is diagnosed simply as
In the previous study using NICAM.12, we often set the value of SICCRT to
300 kg m
Annual mean precipitation during June 2004–May 2005 for the GPCP
product
In the HighResMIP protocol, SST and SIC in the model are fixed to the time-varying boundary conditions. Overall, the global mean impact of the slab ocean model is not very large (circles versus rectangles in Fig. 5). Compared with the fixed SST runs, global mean precipitation and OLR showed slight increases in the slab ocean runs, which are associated with a slightly warmer surface air temperature. In terms of the local climate, however, the fixed SST simulation is known to cause severe bias in the horizontal distribution of clouds and precipitation systems in the tropics (Kodama et al., 2015; Fig. 13), and thus the use of the slab ocean model with a 7 d relaxation time is often preferred. The introduction of the slab ocean model considerably affected the horizontal distribution of clouds and precipitation systems (Fig. 13). A double ITCZ bias was more prominent in the precipitation as well as the high cloud fraction and OLR fields (not shown) in the fixed SST runs compared with the slab ocean runs, particularly in the high-resolution run. Our investigation showed that NICAM16-9S with the slab ocean model best simulated the ITCZ peak precipitation and the precipitation pattern. Although the importance of the short-term SST variation driven by the atmosphere for the precipitation pattern is apparent from Fig. 13, the introduction of the slab ocean model alone does not resolve its bias. Further analysis is necessary to understand the physical mechanisms of the bias, which may be due to factors such as the convective timescale.
No gravity wave drag scheme is used in NICAM.12. In NICAM16-S, the
conventional orographic gravity wave drag scheme
(McFarlane, 1987) is used to better simulate the
location and strength of the subtropical jet. The wave generation parameter
Zonal mean zonal wind (contours) and its bias from JRA-55
reanalysis (shadings) for June–August 2004
Same as Fig. 14 but for the simulation by NICAM16-7S.
Although it is believed that even a mesh size of 14 km is insufficient to
explicitly simulate the effects of the orographic gravity wave drag on the
mean field (Nappo, 2012), introducing such a gravity wave drag
scheme will not necessarily lead to an improvement of the simulated climate
in the global nonhydrostatic model. The gravity wave drag scheme introduces
the parameter
Understanding the dependency of horizontal resolution is a central interest of the HighResMIP. Figure 16 shows the global mean climate in NICAM16-7S (56 km mesh; blue circle), NICAM16-8S (28 km mesh; green circle), and NICAM16-9S (14 km mesh; red circle), along with its sensitivity to the time step of the models' dynamics (including the gravity wave drag scheme) and radiation scheme. Note that the dependency of the time step of the radiation scheme on the global mean climate is negligible.
Same as the left part of Fig. 5 but for the REFFIX, DDT2M, DDT1M, RDT20M, and RDT10M runs.
Global mean precipitation and TOA OLR decreased as the horizontal resolution increased (Fig. 16b and c), consistent with a previous NICAM study using 3.5 to 14 km meshes (Miyakawa and Miura, 2019). The results did not strongly depend on the time step of the model. As seen in Fig. 13, the precipitation pattern in the tropics depends strongly on the resolution: more-dominant double-ITCZ patterns and less-intense local precipitation are simulated as the horizontal resolution is increased. The intense precipitation occurs less frequently in the higher-resolution runs (Fig. 17), consistent with Noda et al. (2012) using older NICAM with 14 and 7 km meshes. The intense precipitation occurs more frequently in the model compared with the Global Precipitation Climatology Project (GPCP) product (Noda et al., 2012; Fig. 17), and it is consistent with the results of Maher et al. (2018), who compared precipitation in global climate models without a convection scheme with precipitation in the GPCP product. Na et al. (2020) showed that the frequency of intense precipitation in the GPCP product is lower than that in the TRMM product, and that the 14 km mesh NICAM without a convection scheme could realistically reproduce the intense precipitation observed by the TRMM satellite (Fig. 17).
Frequency of occurrence (%) of daily mean precipitation binned
with an interval of 1 mm d
The low cloud amount was substantially underestimated, especially in the NICAM16-7S and NICAM16-8S runs (Fig. 16i), leading to an underestimation of the OSR at the TOA (Fig. 16d). Consistently, the net downward shortwave radiation at the surface decreased (Fig. 16k) and the surface air temperature slightly decreased (Fig. 16a) as the horizontal resolution increased. This dependency of the low cloud amount and its related variables, in terms of global mean, on horizontal resolution could be reproduced by changing the time step of the dynamics in the model (the DDT2M and DDT1M runs). Specifically, the time loop in the model is based on the dynamics (Sect. 2.1), and the number of subcycles of the microphysics, turbulence, and surface processes were reduced in the DDT2M and DDT1M runs to keep their time steps unchanged (Table 2). This leads to an increase in frequency of the coupling between the dynamics and these physics schemes. Additional monthly scale sensitivity experiments suggest that the differences between the REFFIX runs and the DDT2M and DDT1M runs are mostly attributed to the altered frequency of physics–dynamics coupling, not the stability issue of the dynamical core (not shown). The low cloud amount was rather greater in the 56 km mesh run than that in the 14 km mesh run under the fixed time step of the dynamics (red circle in the REFFIX run versus blue circle in the DDT1M run in Fig. 16i). Also, the simulated OSR at the TOA is greater and closer to that of the CERES product in the 56 km mesh run compared with the 14 km mesh run with the same temporal resolution, though the better performance in the simulated global mean OSR at the TOA in the 56 km mesh run is a result of a strong compensation between a negative bias off the subtropical west coasts of continents and the southern hemisphere storm-track region and a positive bias in the rest of the lower latitudes (not shown). Such a result of dependency of OSR at the TOA on horizontal resolution with a fixed temporal resolution is similar to that found by Goto et al. (2020), who executed 14 and 56 km mesh online-aerosol NICAM with the same time steps of 60 s for the dynamics, turbulence, and surface schemes and 10 s for the cloud microphysics scheme. Beside the resolution, HighResMIP requires that the model parameters and configuration be the same between the standard and high-resolution runs. However, adapting the time step to the horizontal resolution, as we have done, is a common approach mainly for numerical stability reasons. Therefore, we highlighted here the impact of both the horizontal and temporal resolutions on the global mean climate in the NICAM simulations.
Table 7 describes the computational setting and the simulated year per wall-clock day (SYPD) of the NICAM16-S simulations run on the Earth Simulator 3. The Earth Simulator 3 is an NEC SX-ACE system with 5120 nodes in total for computation, and each computation node has four cores. We often used 10, 40, and 160 computation nodes to run NICAM16-7S, NICAM16-8S, and NICAM-9S, respectively, to obtain a balance between computational efficiency and wall-clock time. An exception was NICAM16-8S for the HighResMIP simulation, for which 160 computation nodes were used to perform a 101-year simulation within a realistic clock time. A file-staging option was used in the NICAM16-8S and NICAM16-9S simulations, whereas the file was directly read and written from a global file system in the NICAM16-7S simulation to reduce queuing time. The actual SYPD was a few times smaller than that shown in Table 7 for NICAM16-8S and NICAM16-9S.
Computational aspects of the simulations on the Earth Simulator 3. They are sampled from 6-month simulations (1 July 2004–31 December 2004).
Figure 18 breaks down the total elapsed time for each component of the model. Unlike a previous evaluation using the K computer (Yashiro et al., 2016), the measurement includes the initial setup and input/output processes. Physics contributes a major part to the total elapsed time. Among the several physics components, the radiation scheme primarily contributes to the total elapsed time, followed by the cloud microphysics scheme, consistent with Yashiro et al. (2016) on the K computer. As the resolution increased, the percentage of time consumed by dynamics increased and the time consumed by the cloud microphysics and turbulence processes decreased because of their invariant time step among the models with different resolutions. An increase in the percentage of the land surface scheme in the simulations with higher resolution and greater number of computational nodes seems to be caused by the node imbalance associated with land–ocean distribution. Because the dynamics involves communication at every time step, some parts of the elapsed time counted as the dynamics can actually be a waiting time due to the node imbalance occurring in other processes.
Percentage of elapsed time for each component of NICAM16-S on the Earth Simulator 3. Times were sampled from 6-month simulations (1 July 2004–31 December 2004). The computational time per 1 d integration is shown on the left.
The data are output on the model's native icosahedral grid on the height above sea level (ASL) or on the standard pressure. The vertical interpolation from the terrain-following height to the ASL height or standard pressure is performed online using second-order Lagrange interpolation during the simulation.
Postprocessing is performed in the following order: All the native icosahedral grid data are converted to high-resolution
latitude–longitude grid data by area–weight averaging. The interval of
latitude and longitude is determined so that the longitudinal interval is
close to the average interval of the icosahedral grid
(Satoh et al., 2014) on the Equator. Specifically,
the interval of longitude and latitude is 0.56 All the high-resolution latitude–longitude grid data are coarsened to
low-resolution latitude–longitude grid data by area–weight averaging. It
is often necessary to reduce the amount of data by such coarsening, although
HighResMIP does not ask us to coarsen the data. We prepared 1.0,
1.25, and 2.5 Several three-dimensional variables are converted from the ASL height to the
standard pressure at this point by linear interpolation, if it is necessary.
After postprocessing steps (1)–(3), monthly mean data are created.
The pressure velocity is obtained from the vertical velocity and temperature
after
This paper describes the experimental design, the latest stable version of
NICAM prepared for the HighResMIP (NICAM16-S), and impacts of NICAM updates
on the simulated climatology using NICAM16-S at different resolutions. The
major updates and their impacts on simulation results are summarized as
follows:
Update of the cloud microphysics scheme: snow and cloud ice were increased,
leading to decreased high cloud amount and increased OLR at the TOA. Implementation of the coupling between cloud microphysics and radiation
schemes: the negative OLR bias was reduced in association with a larger
cloud ice effective radius. Update of treatment of natural and anthropogenic aerosols: the local surface
radiation budget was improved, especially over Africa and South Asia. Update of land surface model: overall, the soil moisture over most of the
Eurasian and the North American continents was increased. Update of the surface albedo values: the cold bias in Greenland, the
Himalayas, and the Antarctic was decreased. Modification of the ICE diagnostics: the warm bias over the Arctic region
was decreased. Introduction of the orographic gravity wave drag scheme: the location and
strength of the zonal mean jet were improved.
Comprehensive evaluations and future projections using full HighResMIP data
by NICAM16-S will be presented in a forthcoming paper.
The exact model source code, input data and scripts to generate them, and
scripts for the simulations and postprocessing used to produce the results
presented in this paper are archived on Zenodo (
CK and ATN managed the overall HighResMIP activity in the NICAM group and prepared the initial and boundary conditions, and MS managed development and scientific activity in the NICAM group. CK added interfaces of the initial and boundary conditions to NICAM. TO added a function to output variables requested by HighResMIP and converted the output data using CMOR3. CK, TO, TS, HY, MN, and YY contributed to the development of NICAM16-S, including debugging, and WS, TN, and DG provided their schemes and/or parameters for the development. CK performed all the HighResMIP simulations and the sensitivity experiments, transferred the data to ESGF, and wrote a major part of this paper. TS wrote most of Sect. 3.1 and 3.2 and TS, ATN, DG, HM, and TN modified the manuscript. All the authors provided advice for the development of NICAM16-S and/or experimental design and reviewed the manuscript.
The authors declare that they have no conflict of interest.
The authors would like to thank Hiroaki Tatebe and Ryosuke Shibuya for discussions on model configurations and Manabu Abe and Takahiro Inoue for technical advice for CMIP6. Chihiro Kodama acknowledges Shunsuke Noguchi for discussions on the vertical resolution of the model. Constructive and careful comments from two anonymous reviewers and Sophie Valcke, the editor in charge, significantly helped improve the manuscript. CERES, CloudSat, ISCCP, GPCP, and TRMM data were obtained from the National Aeronautics and Space Administration (NASA), JRA-55 data from the JMA, and GridSat data from the National Oceanic and Atmospheric Administration (NOAA). The HighResMIP simulations and sensitivity experiments were performed on the Earth Simulator at the Japan Agency for Marine-Earth Science and Technology (JAMSTEC), and some preliminary experiments were performed on the K computer (proposal numbers hp150287, hp160230, hp170234, hp180182, and hp190152).
This study was supported by the Environment Research and Technology Development Fund (JPMEERF20172R01) of the Environmental Restoration and Conservation Agency of Japan (ERCA) and the Integrated Research Program for Advancing Climate Models (TOUGOU) (JPMXD0717935457), the FLAGSHIP2020 project within the priority study4, and Program for Promoting Researches on the Supercomputer Fugaku (Large Ensemble Atmospheric and Environmental Prediction for Disaster Prevention and Mitigation) (JPMXP1020351142) of the Ministry of Education, Culture, Sports, Science and Technology (MEXT) of Japan and JSPS KAKENHI Grant Number JP20H05728.
This paper was edited by Sophie Valcke and reviewed by two anonymous referees.