A global cloud-resolving simulation: Preliminary results from an aqua planet experiment



[1] Results from global simulations using a nonhydrostatic icosahedral-grid AGCM with cloud-resolving resolutions on an aqua planet are discussed. Results depend on the resolution. Simulations with grid intervals of 7 km and 3.5 km include many realistic features in the tropics: hierarchical cloud structures, a Madden-Julian Oscillation (MJO)-like intraseasonal oscillation, and diurnal precipitation cycles. Global cloud-resolving simulations show promise for future climate research. Such models avoid the liabilities associated with cumulus parameterization.

1. Introduction

[2] Atmospheric general circulation models (AGCMs) typically include a cumulus parameterization to represent effects of sub-grid scale cloud convection. However, such parameterizations are based on idealized statistical assumptions and therefore introduce uncertainties into AGCM output. These uncertainties are reduced in “the cloud-resolving convection parameterization” (or “super-parameterization”) [Grabowski, 2001] that embeds a two-dimensional (2D) cloud-resolving model (CRM) at each grid box, which interacts with large-scale motions in the AGCM [Randall et al., 2003; Khairoutdinov and Randall, 2001]. Problems remaining include arbitrary configurations of 2D-CRMs and artificial scale separation between the 2D-CRMs and the host AGCM.

[3] A different approach applies cloud-resolving techniques to the entire globe three-dimensionally (3D), namely, a global cloud-resolving simulation. This approach could improve AGCM results by explicitly calculating multi-scale interactions between cloud-scale and large-scale motions with no artificial scale separation or multi-physical interactions such as cloud-radiation feedback. Cloud-resolving numerical simulations over large horizontal domains have thus far been limited to 2D geometry. For example, Grabowski and Moncrieff [2001, 2002] investigated the longitudinal large-scale organization of convection; Liu and Moncrieff [2004] discussed the formation of the intertropical convergence zone (ITCZ).

[4] Global cloud-resolving simulations have been difficult historically because of limitations in computational resources. Recent advances allow such simulations. In this paper, a newly developed global nonhydrostatic model is run at resolutions that resolve clouds on an aqua planet that approximates an idealized ocean-covered earth. This is the first attempt to simulate the global atmosphere using a 3D cloud-resolving model. Particular focus is given to the tropics, where scale interactions between large-scale circulations and individual clouds in the model output were studied and compared with observations.

2. Model and Experimental Setup

[5] The model used in this study was the Nonhydrostatic ICosahedral Atmospheric Model (NICAM) developed at the Frontier Research Center for Global Change (FRCGC). The dynamical core of the model [Tomita and Satoh, 2004] is based on a nonhydrostatic scheme that is suitable for climate study [Satoh, 2003]. The horizontal grid is a modified icosahedral grid [Tomita et al., 2002]. Horizontal grid intervals for this study were about 14, 7, and 3.5 km, hereafter referred to as Exp-14 km, Exp-7 km, and Exp-3.5 km, respectively. The vertical domain, with a top at 40 km, was comprised of 54 layers with thicknesses that increased from 75 m at the lowest level to 750 m in the upper troposphere. Time steps were 30 sec for Exp-14 km and Exp-7 km and 15 sec for Exp-3.5 km. Cloud microphysics followed a simple scheme that included ice phase effects [Grabowski, 1998]. The same microphysics scheme was applied without cumulus parameterization in all runs. Exp-14 km might not be valid if cumulus parameterizations are not used. The run is included in this study to isolate the impact of resolution in the three runs. A similar approach was used by Juang and Arakawa [2004]. The Mellor and Yamada [1974] level-2 closure scheme for turbulence was used. Surface flux was approximated by a bulk method [Louis, 1979], and radiation was approximated using the two-stream adding method [Nakajima et al., 2000]. Physical processes were calculated at each time step except for radiation. Radiation was calculated every 10 min for Exp-14 km and Exp-7 km and every 5 min for Exp-3.5 km. These radiation time steps are sufficient for cloud-radiation interactions because cloud lifetime is about an hour.

[6] The aqua planet setup was proposed by Neale and Hoskins [2000a]. SST distribution (control case), ozone distribution, and radiative forcing (fixed equinoctial solar incident with a diurnal cycle) are specified. Appropriate initial conditions for NICAM were obtained from a 3.5-year integration from a conventional AGCM, i.e., CCSR/NIES/FRCGC AGCM version 5.7 [Numaguti et al., 1997; K-1 Model Developers, 2004] with T42L59 resolution, an AGCM that has been developed collaboratively by the Center of Climate System Research (CCSR) at the University of Tokyo, the National Institute of Environmental Studies (NIES) and FRCGC. Hereafter, this run will be called Exp-T42. Zonal climatology from the last 3 years of Exp-T42 interpolated onto Exp-14 km gridpoints served as the initial conditions. From those initial conditions, a 90-day integration was run as Exp-14 km. The first 60 days were spin-up; the last 30 days were used to analyze results. The results from Exp-14 km were interpolated at day 60 onto Exp-7 km gridpoints, and a 30-day integration was performed. Similarly, results from Exp-7 km at day 80 served as initial conditions for a 10-day integration for Exp-3.5 km.

3. Results

[7] Before showing detail results, the climatology obtained in this experiment should be validated. The zonal-mean zonal winds and temperatures obtained in Exp-14 km, Exp-7 km and Exp-3.5 km have distributions and intensities that resemble those of Neale and Hoskins [2000b]. With regards to cloud field, the instantaneous albedo in the deep cloud area is over 0.7, while the global albedo during the analyzed term is about 0.2. Although the latter seems to be slightly small in comparison with observations, the direct comparison with observations is not needed to be strict because of the highly idealistic setup of this experiment.

[8] Figure 1 shows outgoing long wave radiation (OLR) averaged over 90 min at day 85 for Exp-3.5 km, and clearly captures the multi-scale cloud structures in the tropics. Figure 1 also includes cloud-free areas in the sub-tropics and the fine structures in baroclinically-driven mid-latitude cyclones with sharp fronts. Figure 2a focuses on OLR between 12°S–12°N and 0°–80°E, illustrating a typical super cloud cluster that extends about 5000 km longitudinally and about 1500 km latitudinally. Furthermore, many cloud clusters with horizontal scales of several hundred kilometers exist within the super cloud cluster. Figure 2b shows surface low and high pressure east and west of the convective region, respectively. Strong westerly winds (a westerly wind burst) are on the western side of the convection, consistent with a convectively coupled Kelvin wave structure. Enhanced easterly winds occur in the upper troposphere (not shown) above the westerly wind burst, consistent with the large-scale organization of a Madden-Julian Oscillation (MJO)-like structure as shown in Figure 6a of Moncrieff [2004].

Figure 1.

Outgoing long wave radiation at day 85 (90 min average) for Exp-3.5 km global run.

Figure 2.

(a) Same as Figure 1 but in 12°S–12°N and 0°–80°E. (b) Zonal velocity at z = 1.5 km (color fill) and surface pressure (contour). for Exp-3.5 km global run.

[9] Figures 3a–3c show Hovmöller diagrams of OLR averaged between 2°S and 2°N during the last 30 days of Exp-14 km and Exp-7 km and during the last 10 days of Exp-3.5 km, respectively. Figure 3 clearly shows cloud clusters with lifetimes of about 2 days moving westward. A few super cloud clusters that act as envelopes of the westward-propagating cloud clusters propagate eastward and are indicated as black lines in Figures 3a and 3b. Such eastward propagation of super cloud clusters and westward propagation of cloud clusters are qualitatively consistent with observations [e.g., Nakazawa, 1988]. However, the phase speed of the super cloud clusters in Exp-14 km is faster than observed: observations suggest that it takes 30–50 days to propagate around the equator, but only 20–25 days are needed in Exp-14 km. Phase speeds in Exp-7 km and −3.5 km are slower than those in Exp-14 km and are closer to observations. For example, the super cloud clusters A, B, and C in Figure 3b show periods of 25-, 30-, and 40-days, respectively. The period of 25 days is short as compared to most observed MJO periods, but such a period is occasionally observed (e.g. ECMWF reanalysis data [Hayashi and Golder, 1993]). The cloud-free area propagates westward (white lines in Figures 3a and 3b) at a speed that is one-third the propagation speed of the super cloud cluster. This is a Rossby wave response. The propagating dry regions occasionally intercept super cloud clusters.

Figure 3.

Hovmöller diagrams of OLR averaged in 2°S–2°N for (a) Exp-14 km, (b) Exp-7 km, and (c) Exp-3.5 km. (d) The zonal distributions of surface pressure averaged in the last 1 day and 5°S–5°N for Exp-14 km (red), −7 km (green), and −3.5 km (blue).

[10] The super cloud clusters in Exp-7 km and Exp-3.5 km show more large-scale organization than those in Exp-14 km. This difference is most notable during the last 15 days in Figures 3a and 3b. Figure 3d shows zonal distributions of surface pressure averaged over the last one day and between 5°S–5°N for the three runs. Exp-14 km has a dominant wavenumber three structure, and Exp-7 km and Exp-3.5 km show wavenumber one structures.

[11] Figure 4 shows how zonal-mean mass-weighted temperature, precipitable water, and the precipitation rate depend on resolution. The mass-weighted temperature and precipitable water in the tropics were 256.5 K and 40–50 kg/m−2 for all three simulations. Tao et al. [2001] summarized the relationship between mass-weighted temperature and precipitable water values derived from many cloud-resolving simulations at radiation-convection equilibrium in the tropics and found a linear correlation between the two quantities. The present results very nearly follow that correlation line. Precipitable water values in this study are slightly smaller than observations of 50–55 kg/m−2 during the Tropical Ocean Global Atmosphere Coupled Ocean-Atmosphere Response Experiment (TOGA COARE), but they agree with some observations of zonal-mean precipitable water [cf. Khairoutdinov and Randall, 2001].

Figure 4.

Zonal-mean distributions of mass-weighted temperature (top), precipitable water (middle), and precipitation rate (bottom) for Exp-14 km(red), −7 km (green) −3.5 km (blue).

[12] Significant differences occurred in precipitation rates in the ITCZ between the three runs: as resolution increased, the precipitation rate decreased. Similarly, the evaporation rate in the sub-tropics decreased as resolution increased (not shown). The higher precipitation rates in the coarser resolution model may have occurred as a result of slower convective development in the convergent trade winds; cloud condensation hardly occurred in the coarser resolution model, which led to stronger convergence near the Equator.

[13] Finally, the precipitation rate had a realistic diurnal cycle in the cloud-resolving model. Recent observations [Nesbitt and Zipser, 2003] show a peak in the precipitation rate in the early morning (around 0500LT) over the open oceans. Figure 5 shows histograms of the mean precipitation rate between 10°S–10°N with a local time interval of 3 h for Exp-T42 which used the Arakawa-Schubert cumulus parameterization, Exp-14 km, Exp-7 km and Exp-3.5 km. Exp-T42 has a precipitation peak at midnight. This peak could be postponed to early morning if a convective suppression mechanism were included as shown by Woolnough et al. [2004]. In contrast, precipitation in Exp-14 km peaks in the morning (0600LT–0900LT), whereas Exp-7 km and Exp-3.5 km show precipitation peaks that are more consistent with observations, that is, in the early morning (0300LT–0600LT). Furthermore, the two runs with highest resolutions have weak semi-diurnal cycles and minor precipitation maxima in the afternoon.

Figure 5.

Histograms of precipitation rate in 10°S–10°N with interval of 3 hours.

4. Concluding Remarks

[14] This paper presents preliminary results of a global cloud-resolving simulation on an aqua planet, with special focus on tropical circulations. The modeling strategy ultimately will improve AGCM climate simulations because multi-scale interactions between cloud-scale and large-scale motions and multi-physical interactions, such as the cloud-radiation feedback, are represented explicitly.

[15] The global CRM qualitatively captured hierarchical structures in clouds even at the coarsest resolution (14 km). Simulations with grid intervals of 7 km and 3.5 km reproduced an intraseasonal oscillation similar to the MJO: a wavenumber one structure existed and the phase speed of the convective region compared favorably with observations. The diurnal precipitation cycle was also well represented in the simulations with grid intervals of 7 km and 3.5 km: the peak in precipitation occurred in the early morning, as observed. ITCZ intensity varied with horizontal resolution and numerical results had not yet converged at a horizontal grid interval of 3.5 km. This result occurs because the horizontal grid interval of 7 km is still coarse for calculations that resolve clouds. Results from the 3.5-km run are likely more realistic; the 7-km run may be improved by better modeling of physical processes such as boundary layer and meso-scale circulations. Replacing the plane-parallel radiation processes with 3D radiation processes that consider cloud and topographic shadows will also help to improve physical processes in the high-resolution simulations.

[16] Limitations in computational resources precluded an integration of the global cloud-resolving model for periods longer than a month, which is too short a period to study climatology or seasonal changes. The results in this paper, however, suggest that use of global cloud-resolving models is promising and such models could be useful tools for climate study in the near future.


[17] We thank the model development team of the “K-1 Project” (subject No. 1 of the Kyousei Project/Human-Nature-Earth Symbiosis Project supported by the Research Revolution 2002 Program of the Ministry of Education, Culture, Sports Science, and Technology of Japan), who provided the newest code for the CCSR/NIES/FRCGC AGCM. All calculations for this study took place at the Earth Simulator in the Earth Simulator Center of the Japan Agency for Marine-Earth Science and Technology.