Climatology of a nonhydrostatic global model with explicit cloud processes



[1] This is the first study to obtain a statistical climate state under a perpetual July condition using a nonhydrostatic global model with explicit cloud processes. Horizontal grid intervals of approximately 7 and 14 km over the globe marginally allowed for a general representation of meso-scale circulations associated with deep clouds. The fine-scale precipitation distribution was similar to satellite observations, particularly in regions of the Intertropical Convergence Zone. Realistic distributions of cloud cover in upper, middle, and lower levels were also obtained, although the geographic locations of lower clouds differed slightly from observations over the eastern Pacific. Sensitivity experiments revealed a decrease in outgoing longwave radiation (OLR) as ice fall speed increased or as mixing in the boundary layer was enhanced. As the horizontal resolution increased, the cloud fraction increased and both the OLR and the total water path decreased, implying that cloud overlap also decreased as the resolution increased.

1. Introduction

[2] Clouds play important roles in determining climate. Because of limits to computational resources, most current atmospheric general circulation models (AGCMs) do not explicitly resolve either deep or shallow clouds; instead, they use parameterizations to represent sub-grid-scale cloud convection. These models exhibit a large range of sensitivity in cloud feedbacks [Soden and Held, 2006]. Shallow clouds have been recognized as a major factor determining cloud feedbacks [Bony and Dufresne, 2005]. However, the treatment of deep clouds is also important, and global circulation is sensitive to cumulus parameterization; cumulus parameterization models have difficulties in reproducing realistic behaviors of convective systems such as Madden-Julian oscillation (MJO) [Slingo et al., 1996; Sperber et al., 1997; Waliser et al., 1999]. Our ultimate goal is to improve the statistical properties of convective cloud systems by explicitly calculating cloud processes for mesh intervals of a few kilometers instead of using cumulus parameterizations.

[3] Recently, a series of global atmospheric simulations with explicit cloud processes has been performed using the Nonhydrostatic Icosahedral Atmospheric Model (NICAM) [Tomita and Satoh, 2004; Satoh et al., 2007] for horizontal mesh intervals of Δx ≃ 3.5, 7, and 14 km. Tomita et al. [2005] were the first to perform simulations at these resolutions using an aqua-planet setup. They treated multi-scale and multi-physical interactions explicitly, reproducing the multi-scale organization of convection in tropical super-cloud clusters [Nasuno et al., 2007]. For a case with realistic topography, Miura et al. [2007] performed a short-duration simulation using reanalysis data for the initial condition at 00 UTC 1 April 2004. In their experiment, the evolution of cloud clusters and the development of a typhoon were realistically simulated for the tropics, as well as for mid-latitude cyclones. They also showed that the organization of tropical convection was sensitive to boundary layer processes.

[4] Numerous discussions have addressed the resolution required to represent clouds explicitly. Much of the mesoscale structure is represented for a horizontal grid interval Δx = 4 km [Weisman et al., 1997]. For finer structures such as shallow convection, a much higher resolution is needed. Models of the general circulations that explicitly resolve mesoscale cloud systems within grid intervals of a few kilometers are considered useful. The results of a series of experiments using the nonhydrostatic global model NICAM indicate that for Δx ≃ 3.5, 7, and 14 km, large-scale behaviors were similar in the sense that eastward-propagating organized convective systems were commonly observed at these resolutions [Tomita et al., 2005; Satoh et al., 2005; Nasuno et al., 2007]. The results of Pauluis and Garner [2005] also indicate that the statistical properties of deep convection such as vertical velocity are well represented for Δx ≤ 16 km. Our aim was to examine the global representation of real climatology using explicit cloud processes without any cumulus parameterization for Δx ≃ 7 and 14 km, which are marginal mesoscale-resolving scales.

[5] Using NICAM, we have performed climate experiments with durations of several months, referring to a perpetual July setting [Cess and Potter, 1988]. Here, we compare the simulation results to observations. We also examine the dependencies on the ice fall speed, the turbulent mixing length, and horizontal resolution.

2. Experimental Design

[6] NICAM is based on the icosahedral grid system in which horizontal grids are quasi-uniformly distributed on a sphere [Tomita et al., 2002]. We used the following schemes: a cloud microphysics scheme with simple treatment of ice effects proposed by Grabowski [1998], a bulk-type surface flux scheme described by Louis [1979], and the MSTRNX radiation model [Nakajima et al., 2000; Sekiguchi, 2004]. No statistical cloud-overlapping parameterization is used in the radiation scheme. For the turbulence process [Mellor and Yamada, 1974], a level-2 closure scheme with a simple moisture effect, i.e., the saturated Brunt-Väisälä frequency of Durran and Klemp [1982], is used to estimate the Richardson Number in a saturated grid. For soil processes, the bucket model is used for temperature, and soil water is fixed to the initial value. Sea surface temperature (SST), sea ice, and ozone are fixed to those of the July climatology, and the solar incidence of 15 July is given for each day. The perpetual July setting with fixed SST is convenient because the equilibrium state is reached in a relatively short time. To obtain an initial condition, we first conducted a 12-year simulation using the MIROC model [K-1 Model Developers, 2004] with a bucket model for the resolution of T42L25. We then used the average of July climatology for the last 10 years for the initial condition of the NICAM simulations. The horizontal grid intervals used were approximately Δx ≃ 14 km and Δx ≃ 7 km. These resolutions correspond to grid-recursive division levels (glevel) 9 and 10, respectively [Tomita et al., 2001]. The vertical domain had 40 layers with the model top of z = 40 km.

[7] As described later, the control experiment showed larger outgoing longwave radiation (OLR) than did the observed climatology. To examine the sensitivity of OLR, we varied the turbulence mixing length and snow fall speed. Table 1 provides the parameter sets used. The parameter cs is the same c as that of Grabowski [1998, equation 6b]; the snow fall speed increases as cs increases. L0 is the maximum mixing length, which determines the mixing length LzL0/(z + L0). CS4L100 is the control experiment, and experiments CS4L200, CS3L200, and CS3L200dx7 were used to examine the dependencies described above. For CS4L100, CS4L200, and CS3L200, calculations were performed for 120, 120, and 200 days, respectively, until an equilibrium state was reached. For CS3L200dx7, calculation was performed for 40 days from a spin-up state of 170 days of CS3L200 because of limited computational resources. Although not yet equilibrated, the tendency of OLR discussed below can be considered robust. For each case, the last month is used for analysis.

Table 1. Parameter Settings and 30 Day Mean Globally-Averaged Results
LabelΔx, kmcsL0, mOLR, Wm−2OSRU, Wm−2Cloud FractionTotal Water Path, kgm−2
July climatology (ERBE)   239.396.0

3. Climatology of the Control Experiment

[8] The 30-day averaged zonal mean temperature, zonal velocity, and water vapor for the control case (CS4L100) were compared for their differences from the July climatology of ERA-40 (European Center for Medium-range Weather Forecasting [ECMWF] reanalysis data [Kallberg et al., 2004]; see Figure 1). The obtained statistics were similar to the observed climatology. The temperature distribution was close to that of ERA-40, except at high latitudes, where it was higher by a few degrees. The extratropical jet in the Southern Hemisphere and the equatorial easterly winds were both slightly stronger for CS4L100 than for the observations, especially in the stratosphere. This difference may reflect the lack of gravity wave parameterization in our model. The water vapor distribution in lower troposphere in the Northern Hemisphere had a wet bias up to 0.002 kg/kg. This is presumably caused by the perpetual July setting in which the soil was permanently heated by the strong solar incidence in the Northern Hemisphere, whereas soil moisture was fixed to the climatology. These tendencies have also been found in a perpetual July experiment using the MIROC model and a different representation of soil processes (MATSIRO [Takata et al., 2003]). In 30°S–0°, water vapor distribution had wet bias up to 0.001 kg/kg near boundary layer and dry bias up to 0.001 kg/kg above boundary layer. This tendency is quantitatively very similar to the results for Δx = 16 km found by Pauluis and Garner [2005] in which they discuss that the bias is due to the coarse resolution insufficient to represent mixing by shallow clouds. This suggests that the bias in our results would decay as resolution increases.

Figure 1.

The 30-day averaged zonal mean temperature, zonal velocity, and water vapor for CS4L100. Color shading indicates the difference from the July climatology of ERA-40 [Kallberg et al., 2004].

[9] Figure 2 compares the 1-month averaged precipitation of CS4L100 with that of July 2006 obtained by the 3B43 Tropical Rainfall Measuring Mission (TRMM) combined with other data sources. Overviews of July precipitation were similar for CS4L100 and TRMM. In finer detail, in the ITCZ and South Asian regions, the fine-scale structure of simulated precipitation resembled that of the TRMM, whereas there were precipitation areas southeast of Japan and in Indonesia that were not shown in the TRMM climatology for July. In Figure 3a, zonal mean precipitation is compared with Climate Predication Center (CPC) Merged Analysis of Precipitation (CMAP) data [Xie and Arkin, 1997]. At low latitudes, the obtained precipitation was similar to the CMAP values (within 10%), whereas it was larger by 50% at intermediate and high latitudes. The bias in the northern extratropics may be associated with the warm bias, although CMAP data are less accurate at higher latitudes.

Figure 2.

(a) The 30-day averaged precipitation (mm/month) for CS4L100. (b) Precipitation of July 2006 from 3B43 Tropical Rainfall Measuring Mission (TRMM) and other data sources.

Figure 3.

(a) Zonal mean precipitation (30-day average) for each simulation and for the July climatology during 1979–2003 by CMAP. (b) Zonal mean outgoing longwave radiation (OLR, 30-day average) for each simulation and for the July climatology by ERBE.

[10] In Figure 4, the cloud distribution calculated using an International Satellite Cloud Climatology Project (ISCCP) simulator is compared with that of ISCCP cloud data [Rossow and Schiffer, 1999]. These distributions are quite similar. In detail, the cloud fraction for the high level is larger for CS4L100, but smaller for the middle level. The shallow cloud distribution is also comparable to the observed distribution, although clouds off the west coast (California and Peru) are insufficient. This result is encouraging because no sub-grid parameterization for shallow clouds was used in this simulation.

Figure 4.

Cloud distribution categorized by the cloud-top height (high, 0–440 hPa; middle, 440–680 hPa; low, <680 hPa). (top) Values obtained by the ISCCP simulator for CS4L100 and (bottom) July mean ISCCP cloud data during 1986–1999.

4. Sensitivity Study

[11] For the climate study, radiative flux is one of the concern. In Table 1, globally averaged OLR and upward outgoing short wave radiation (OSRU) for each experiments are compared with the July climatology by Earth Radiation Budget Experiment (ERBE) S-9 data [Harrison et al., 1990]. OLR for the control experiment (CS4L100) was 16 Wm−2 larger than that for ERBE. This large positive bias can be explained by the cloud distribution because the OLR bias for clear sky was small (not shown). Next, we examined the sensitivities of the snow fall speed (cs) and horizontal resolution. The turbulence mixing length was also changed to transport water vapor to higher levels because lower levels had more water vapor than upper levels compared with the ERA-40 data. Table 1 shows that on average, OLR decreased as (1) the mixing length increased (L0 = 100 to 200), (2) the snow fall speed decreased (cs = 4 to 3), or (3) the resolution increased (Δx ≃ 14 km to 7 km). The dependency on cs is reasonable because as the snow fall speed decreases, the anvils associated with deep clouds become wider and hence reduce OLR. The dependency on the boundary layer scheme is rather notable; as vertical mixing was enhanced, the moisture in the free troposphere became more abundant and affected OLR. This dependency on boundary layer mixing is consistent with results by Miura et al. [2007]. Although CS3L200 OLR on average were closest to ERBE values among our experiments, meridional OLR profiles in Figure 3b shows that the CS4L100 meridional OLR profile had the most similar shape to that of ERBE. Correspondingly, cloud profile of CS4L100 was the most similar to that of ISCCP (not shown for other parameters). For CS3L200 and CS3L200dx7, halving the horizontal grid interval decreased OLR by 10 Wm−2. These fact indicate that the bias in OLR for CS4L100 decreases if resolution increases. As for OSRU, Table 1 shows that biases were within 4Wm−2 for all experiments.

[12] In Table 1, cloud fraction and total water path are also compared. The cloud fraction of CS3L200dx7 was larger than that of CS3L200, which would be a cause of the smaller OLR. In contrast, the total water path of CS3L200dx7 was smaller than that of CS3L200. These results indicate that cloud overlap decreases as the resolution increases at low latitudes. One explanation for this finding is that cloud overlap decreases because of the increased degree of freedom of overlapping. Each cloud size also became smaller as Δx became smaller.

5. Summary

[13] We performed climate simulations with explicit cloud processes using a perpetual July condition with fixed SST for Δx ≃7 km and 14 km. For the control experiment (Δx ≃ 14 km), the overview of climatology was similar to observations. The fine-scale structure of precipitation was comparable to that of TRMM. Cloud cover in the upper, middle, and lower levels were also similar to the ISCCP clouds. These results indicate that global-scale climatology can be represented using the 14-km mesh global nonhydrostatic model, although any cloud parameterization is not used for the mesh size which is marginal to represent cloud system. Nonetheless, the resolution dependency shows that halving Δx to 7 km decreased OLR by 10 Wm−2, indicating that the treatment of the smaller scale affects the radiation budget. The resolution dependency also shows that as the resolution increases, the cloud fraction increases and the total water path decreases, implying that the cloud overlap decreases as horizontal resolution increases. This may be because of the increased degree of freedom of overlapping.

[14] Although our experiments with explicit cloud processes were free from the ambiguity of cumulus parameterization, ambiguities remained with other parameters. Sensitivity experiments show that both the ice fall speed and mixing length affect OLR. As the ice fall speed decreases, OLR is reduced as the anvils associated with deep clouds become wider. As vertical mixing is enhanced, moisture in the free troposphere becomes more abundant and affects OLR. These sensitivities will be described in detail elsewhere.


[15] All of the simulations performed in this study were done using the Earth Simulator at the Japan Agency for Marine-Earth Science and Technology. This research was supported by the Core Research for Evolutional Science and Technology (CREST) program of the Japan Science and Technology Corporation (JST). We thank H. Miura and T. Nasuno for helpful discussions that helped us to improve this manuscript.