An ocean-atmosphere climate simulation with an embedded

cloud resolving model

Cristiana Stan,1 Marat Khairoutdinov,2 Charlotte A. DeMott,3 V. Krishnamurthy,1,4

David M. Straus,1,4 David A. Randall,3 James L. Kinter III,1,4 and J. Shukla4,5

Received 4 September 2009; revised 20 November 2009; accepted 30 November 2009; published 5 January 2010.

[1] Mean climate and intraseasonal to interannualvariability of two versions of the Community ClimateSystem Model (CCSM) coupled atmosphere-ocean generalcirculation model (CGCM) are analyzed. The first version isthe standard CCSM, in which cloud effects on the large-scalecirculation are represented via parameterizations. The secondversion includes ‘‘super-parameterization’’ (SP) of convectiveprocesses by replacing parameterized cloud processes with atwo-dimensional (2D) cloud-process resolving model (CRM)at each CGCM grid column. The SP-CCSM improves severalshortcomings of the CCSM simulation, including meanprecipitation patterns, equatorial SST cold tongue structureand associated double intertropical convergence zone (ITCZ),the Asian monsoon, periodicity of the El Nino–SouthernOscillation, and the intraseasonal Madden-Julian Oscillation.These improvements were obtained without the retuning ofthe coupled model, which is surprising in view of previousexperience with other coupled models. Citation: Stan, C.,

M. Khairoutdinov, C. A. DeMott, V. Krishnamurthy, D. M.

Straus, D. A. Randall, J. L. Kinter III, and J. Shukla (2010), An

ocean-atmosphere climate simulation with an embedded

cloud resolving model, Geophys. Res. Lett., 37, L01702,

doi:10.1029/2009GL040822.

1. Introduction

[2] The horizontal resolution in current coupled ocean-atmosphere general circulationmodels (CGCMs) is truncatedat a point in the energy spectrum that does not allow theexplicit representation of important physical processes suchas cloud formation, turbulent mixing in the boundary layer,eddy mixing in the ocean, and the floe-scale physics of sea-ice. Themodels include parameterizations of these processes,which lead to large errors and uncertainties in the results.[3] The latest assessment of climate model simulations

and projections using state-of-the-art CGCMs, issued by theIntergovernmental Panel on Climate Change (IPCC), reiter-ates the conclusion of earlier reports that clouds still representthe largest source of uncertainty in climate simulations[Randall et al., 2007]. While cloud parameterizations havebeen improved for over half of a century [Randall et al.,

2003], the systematic errors associated with the shortcomingsof the parameterizations remain large. These range fromerrors in the simulated background climate to errors in thefrequency and intensity of internal fluctuations on variousspace and time scales [Guilyardi et al., 2009].[4] We demonstrate that replacing the parameterization of

cloud processes with a limited-domain, 2D CRM improvescoupled global climate simulation on a wide range of spatialand temporal scales.

2. Model Description

[5] We use a state-of-the-art version of the CommunityClimate System Model, version 3 (CCSM) [Collins et al.,2006], with a 2D CRM embedded in each CGCM atmo-spheric grid column, through a multi-scale modeling frame-work (MMF) [Grabowski, 2001; Khairoutdinov andRandall, 2001;Khairoutdinov et al., 2005]. TheMMF allowssimultaneous representations of the large-scale atmosphericcirculation on the coarse-resolution grid (T42) of the atmo-spheric component of the CGCM and physical processessuch as convection and stratiform cloudiness on the fine-resolution grid (4 km) of the CRM. Cloud microphysics andradiation are parameterized, but applied at the CRM scale. Aswith traditional parameterizations, CRM results are assumedto be representative of cloud processes taking place within aGCM grid column, and not exact representations of thoseclouds. Convective tendencies of heat, moisture, and mo-mentum are communicated to the large scale via the GCMand, as such, do not propagate from one CRM to another,except through their effects on the large-scale environment.[6] The embedded CRMs are sometimes referred to as

super-parameterization (SP) of small-scale processes. Thecoupled model with a CRM in each grid column of theatmospheric component will hereafter be referred to asSP-CCSM. While the CRM is a 2D model described byKhairoutdinov and Randall [2001, 2003], the SP-CCSMincludes a wide range of spatial and temporal scales andtheir interactions. Observational evidence suggests thatsuch interactions play a key role in the spatiotemporalorganization of tropical convection [e.g., Slingo et al.,2003; Zelinka and Hartmann, 2009].[7] We assessed the implication of the cloud processes

representation by comparing the climate simulation pro-duced by the SP-CCSM with the observations and also witha control simulation obtained from a version of the CCSMthat uses parameterized cloud processes.

3. Results

[8] Figure 1 compares the monthly Nino-3.4 time seriesof the sea surface temperature (SST) anomalies associated

GEOPHYSICAL RESEARCH LETTERS, VOL. 37, L01702, doi:10.1029/2009GL040822, 2010ClickHere

for

FullArticle

1Center for Ocean-Land-Atmosphere Studies, Calverton, Maryland, USA.2School of Marine and Atmospheric Sciences, Stony Brook University,

Stony Brook, New York, USA.3Department of Atmospheric Science, Colorado State University, Fort

Collins, Colorado, USA.4Department of Atmospheric, Oceanic and Earth Sciences, George

Mason University, Fairfax, Virginia, USA.5Institute of Global Environment and Society, Calverton, Maryland, USA.

Copyright 2010 by the American Geophysical Union.0094-8276/10/2009GL040822$05.00

L01702 1 of 6

with the El Nino–Southern Oscillation (ENSO) as simulatedby the SP-CCSM (1b) and CCSM (1c), and as observed (1a)based on the Hadley Centre SST version 2 (HadSST2) dataset. The SP-CCSM produces a sequence of ENSO events thatis less regular than in the control simulation. The irregularityof events is also notable in the broadening of the main peak ofthe power spectrum (Figure 1d). The amplitude of the signalis more realistic, but the period of the spectral peak in theoscillation is too short by 18 months when compared to theobservations. However, the period of ENSO simulated bythe SP-CCSM has improved compared to the CCSMsimulation. We note that ENSO is a coupled phenomenondriven by the interactions between the ocean and theatmosphere. The improvement in the period and irregu-larity in the SP-CCSM is solely due to a better represen-tation of clouds in the atmospheric model. The resolutionof the ocean model can also be a key factor [Roberts et al.,2009]; the resolution used in our simulation (gx3) isknown to be less ideal for ENSO simulation [Timmermannet al., 1999].

[9] We find that the improvement in ENSO simulation isconsistent with the SP-CCSM’s ability to capture the correctstructure of the equatorial cold tongue, which in the CCSMis too narrowly confined to the equator and extends too farinto the western tropical Pacific (not shown). The SP-CCSMalso simulates the opposite relation with the Nino-3.4 regionacross the central and western Pacific, although it is not asstrong as observed. The complex system of islands in thisregion influences the wind-stress simulation. Increased res-olution in all components of the CGCM will allow a betterrepresentation of the wind-stress acting at the air-sea interfaceand over the Maritime Continent [Shaffrey et al., 2009].[10] Realistic simulation of precipitation is one of the

biggest challenges for the current generation of models.Comparing the global distribution of the seasonal meanrainfall simulated by SP-CCSM (Figure 2b) and CCSM(Figure 2c) with observations (Figure 2a) taken from theClimate Prediction Center (CPC) Merged Analysis Precip-itation (CMAP) dataset, the improvement in SP-CCSM isapparent in many regions. In both boreal winter and summer

Figure 1. Time series of the Nino 3.4 index and the associated power spectrum. (a) Observations (HadSST; the SSTanomaly is computed with respect to the 1959–2002 climatology and only the last 22 years are shown to be consistent withthe length of the model time series), (b) SP-CCSM, and (c) CCSM simulations for 22 years. (d) Power spectra computedfrom the time-series in Figures 1a–1c. The dashed line represents the red noise power spectrum.

L01702 STAN ET AL.: A GLOBAL CLIMATE SIMULATION L01702

2 of 6

seasons, the distribution of rainfall along the inter-tropicalconvergence zone (ITCZ) and South Pacific convergencezone (SPCZ) in the SP-CCSM reproduces the observedpatterns, and does not produce the spurious double ITCZ

present in the CCSM. However, the SP-CCSM’s rainfalleast of the date line in the SPCZ does not extend into theextratropics as in the observations. During the summer,the distribution of precipitation over the western part of

Figure 2. Seasonal mean precipitation and root mean square error (RMSE) between 60�S and 60�N. For regions markedwith boxes the RMSE was computed (see text for details). (a) Observed December–January–February (DJF) 1979–2006mean from CMAP. (b) As Figure 2a but for June–July–August (JJA). (c) DJF mean from SP-CCSM. (d) As Figure 2c butfor JJA. (e) DJF CCSM. (f) As Figure 2e but for JJA. (g) DJF mean for 1986–2003 from SP-CAM. (h) As Figure 2g but forJJA. Units are in mm/day.

L01702 STAN ET AL.: A GLOBAL CLIMATE SIMULATION L01702

3 of 6

the tropical ocean basin shows an improvement in theSP-CCSM, but it is still weak compared to the CMAPanalysis. We find a significant improvement in the simu-lation of rainfall over the South Asian monsoon region,where large-scale model biases have been reduced. Tworealistically simulated maxima, one west of the Indianpeninsula and the other in the Bay of Bengal, are aresponse to moisture advection by the onshore flow.Compared to observations, the precipitation variation overIndia is less well simulated, possibly due to the inability ofcoarse resolution of the CGCM to resolve orography[Gent et al., 2009]. However, compared with the CCSM,we see a significant improvement in the simulation ofmonsoon variability such as the interannual variability ofthe summer monsoon associated with the SST anomalies(not shown). Regional improvements are quantified withRMSE for specific regions marked by boxes in Figure 2 inTables S1 and S2.1

[11] Results by Woolnough et al. [2000] and Manganelloand Huang [2009] have shown that biases in the tropicalSST simulations of coupled models impact the simulation ofthe rainfall distribution and its variability. The tropicalatmosphere is very sensitive to changes in the SST, and smallSSTanomalies can lead to large changes in the distribution ofconvection [Trenberth et al., 1998]. For comparison, asimulation with the atmospheric component of SP-CCSM(SP-CAM) using specified (observed) SSTand sea-ice distri-butions as boundary conditions [Khairoutdinov et al., 2008]was analyzed. Figures 2g and 2h show thewinter and summerrainfall distributions simulated by SP-CAM. In both seasons,there are regions with excessive precipitation compared toobservations. The most notable biases are found during thesummer over the western Pacific Ocean and the Asianmonsoon region. In the SP-CCSM, these biases are greatlyalleviated, and as a result, the coupled model produces abetter simulation of the geographical distribution and ampli-tude of precipitation than the uncoupled model does. Thisfinding is gratifying but surprising, since coupling to anocean model typically makes the results of an atmospheremodel worse [Hurrell et al., 2006]. Comparison of thesimulated and observed SST fields are provided in theauxiliary material. Tables S1 and S2 show the associatederrors in the SST for regions marked with boxes in Figure 2.[12] The impact of CRMs on the simulation of tropical

variability discussed in connection with SST and precipita-tion can also be seen in the simulation of the dominantmode of intraseasonal variability of the tropical atmosphere,the Madden-Julian Oscillation (MJO), which is poorly sim-ulated in most of the current generation of models [Randallet al., 2007]. The phase composites of the dominant MJOoscillatory mode extracted using a multi-channel singularspectrum analysis (MSSA) [Ghil et al., 2002;Krishnamurthyand Shukla, 2007] of outgoing longwave radiation (OLR)anomalies in observations and simulations are compared inFigure 3. The composites are constructed by averaging theOLRMJOmode anomalies (reconstructed components) overseveral equal intervals in a (0, 2p) phase cycle of the MJOoscillation for the entire analyzed period. The MJO-relatedmode of variability simulated by the SP-CCSM occurs atrealistic period (56 days in SP-CCSM and 52 days in theobservations) and propagates eastward with a realistic phasespeed. The CCSM produces MJO-like variability but theperiod of oscillation is much longer (72 days) and thedisturbance propagates more slowly. Analysis of the spatialstructure of the MJO during its average lifecycle in CCSM(Figure 4) reveals that the parameterization of convectionresults in weak convective anomalies in the Indian Ocean thatcannot propagate beyond the Maritime Continent and decaysbefore reaching the central Pacific Ocean. In the SP-CCSM,the alternating ‘‘active’’ periods of enhanced convection and‘‘break’’ periods of reduced convection over the IndianOcean are in agreement with the observations, as are theireastward and meridional propagation.[13] Complementary to the MSSA analysis, we have also

applied the CLIVAR [CLIVAR Madden-Julian OscillationGroup, 2009] standardized set of MJO diagnostics (notshown). We find that the CCSM has additional shortcom-ings in representing MJO characteristics such as weaknorthward propagation during the boreal summer and poorrepresentation of the lag of zonal wind anomaly behind (to

Figure 3. Phase composites of OLR MJO mode recon-structed component (RC) for CCSM (blue), SP-CCSM(red), and NOAA observation (green). The composites areaverages of RCs over phase intervals of length p/36, p/28,and p/26 of the MJO oscillations of CCSM, SP-CCSM, andNOAA observation, respectively. The choice of the phaseinterval was based on the average period of the MJOoscillation in the models and observation (72, 56, and 52days, respectively) such that each phase interval is 2 dayslong. The composites are plotted by averaging over (5�S–5�N) at (top) 60�E and (bottom) 120�E. Day 0 correspondsto the composite over the phase interval 0–p/36, 0–p/28,and 0–p/26 for CCSM (blue), SP-CCSM (red), and NOAAobservation (green), respectively, and represents the com-posites for the next 2 days of the average MJO oscillation.The sequence follows the phase intervals in the (0–2p)cycle of the oscillation’s phase. Units are W m�2.

1Auxiliary materials are available in the HTML. doi:10.1029/2009GL040822.

L01702 STAN ET AL.: A GLOBAL CLIMATE SIMULATION L01702

4 of 6

the west of) the precipitation peak. The SP-CCSM improvesthe simulation of these processes.

4. Discussions and Conclusions

[14] Results reported here provide compelling evidencethat replacing the cloud process parameterizations withembedded CRMs improves some of the known shortcom-ings of the CCSM. Since these shortcomings affect most ofthe current CGCMs we argue that currently availableparameterizations for representing the overall properties ofclouds in a grid box have deficiencies that limit the modelskill in simulating the observed characteristics of the Earth’sclimate and its variability.[15] Convection and cloud parameterizations of most

atmospheric general circulation models (AGCMs) havebeen tuned in uncoupled simulations to improve the modelresults, which in some cases may be achieved by thecancelation of errors. For example, errors in a cloud micro-physics parameterization can be tuned to cancel errors inthe convection parameterization. When the AGCM iscoupled to an ocean model, the errors do not cancel any

more, and the AGCM results usually become less realistic.As shown above, when the SP-CAM is coupled to an oceanmodel, its results actually become more realistic, counter toexpectations based on prior experience with parameterizedmodels.[16] Two potential explanations come to mind. First,

improvements may arise simply because ocean-atmosphereinteraction on short time scales plays a role in determiningthe observed climate. Second, conventional convectiveparameterizations are based on the assumption that collec-tive effects of cloud ensembles are in quasi-equilibrium withthe large-scale forcing [Arakawa and Schubert, 1974]. Thisapproximation affects a conventionally-parameterized mod-el’s ability to simulate the stochastic variability of theobserved climate that arises from internal high-frequencyfluctuations [Neelin and Zeng, 2000]. Observational studies[e.g., McPhaden et al., 2006; Hendon et al., 2007] haveshown that ENSO is affected by the high-frequency vari-ability of westerly wind anomalies in the western Pacificduring MJO events. Our results suggest that improvementsof the ENSO variability in the SP-CCSM could be linked, atleast in part, to the model’s ability to simulate the MJO.

Figure 4. Phase composites of the dominant OLR MJO mode reconstructed component for 2-day long phase intervals for(left) CCSM, (middle) SP-CCSM, and (right) NOAA observation. The sequence follows the phase intervals in the (0–2p)cycle of the oscillation’s phase. The average period of one cycle of these eight phase intervals is 72 days for CCSM, 56days for SP-CCSM, and 52 days for NOAA observation. Units are W m�2.

L01702 STAN ET AL.: A GLOBAL CLIMATE SIMULATION L01702

5 of 6

[17] This experiment suggests that MMF approachshould not be limited to the atmospheric model but extendedto the other components of the climate model, which also useparameterizations of small-scale process. These results alsobode well for the ability of the global cloud resolving modelsto simulate Earth’s climate.

[18] Acknowledgments. We acknowledge the support of the Com-putational and Information Systems Laboratory at NCAR for providingcomputer time for this work. A portion of this work has been supported bythe NSF Science and Technology Center for Multi-Scale Modeling ofAtmospheric Processes, managed by Colorado State University undercooperative agreement ATM-0425247. C. S., V. K., D. M. S., J. L. K., andJ. S. were supported by the NSF (OCI-0749290 and ATM-0332910), NOAA(NA04OAR4310034), and NASA (NNG06GB54G and NNG04GG46G).

ReferencesArakawa, A., and W. H. Schubert (1974), Interaction of a cumulus cloudensemble with the large-scale environment, part I, J. Atmos. Sci., 31,674–701.

CLIVAR Madden-Julian Oscillation Group (2009), MJO simulation diag-nostics, J. Clim., 22, 3006–3030.

Collins, W. D., et al. (2006), The Community Climate System Modelversion 3 (CCSM3), J. Clim., 19, 2122–2143.

Gent, P. R., S. G. Yeager, R. B. Neale, S. Levis, and D. A. Bailey (2009),Improvements in a half degree atmosphere/land version of the CCSM,Clim. Dyn., doi:10.1007/s00382-009-0614-8, in press.

Ghil, M., et al. (2002), Advanced spectral methods for climatic time series,Rev. Geophys., 40(1), 1003, doi:10.1029/2000RG000092.

Grabowski, W. W. (2001), Coupling cloud processes with the large-scaledynamics using the cloud-resolving convection parameterization(CRCP), J. Atmos. Sci., 58, 978–997.

Guilyardi, E., A. Wittenberg, A. Fedorov, M. Collins, C. Wang, A. Capotondi,G. J. van Oldenborgh, and T. Stockdale (2009), Understanding El Nino inocean-atmosphere general circulation models: Progress and challenges,Bull. Am. Meteorol. Soc., 90, 325–340.

Hendon, H. H., M. C. Wheeler, and C. Zhang (2007), Seasonal dependenceof the MJO-ENSO relationship, J. Clim., 20, 531–543.

Hurrell, J. W., J. J. Hack, A. S. Phillips, J. Caron, and J. Yin (2006), Thedynamical simulation of the Community Atmosphere Model version 3(CAM3), J. Clim., 19, 2162–2183.

Khairoutdinov, M., and D. A. Randall (2001), A cloud resolving model as acloud parameterization in the NCAR Community Climate System Model:Preliminary results, Geophys. Res. Lett., 28, 3617–3620.

Khairoutdinov, M., and D. A. Randall (2003), Cloud-resolving modeling ofARM Summer 1997 IOP: Model formulation, results, uncertainties andsensitivities, J. Atmos. Sci., 60, 607–625.

Khairoutdinov, M., D. A. Randall, and C. A. DeMott (2005), Simulationof the atmospheric general circulation using a cloud-resolving modelas super-parameterization of physical processes, J. Atmos. Sci., 62,2136–2154.

Khairoutdinov, M., C. A. DeMott, and D. A. Randall (2008), Evaluation ofthe simulated interannual and subseasonal variability in an AMIP-stylesimulation using the CSU multiscale modeling framework, J. Clim., 21,413–431.

Krishnamurthy, V., and J. S. Shukla (2007), Intraseasonal and seasonallypersisting patterns of Indian monsoon rainfall, J. Clim., 20, 3–20.

Manganello, J., and B. Huang (2009), The influence of systematic errors inthe southeast Pacific on ENSO variability and prediction in a coupledGCM, Clim. Dyn., 32, 1015–1034.

McPhaden, M. J., X. Zhang, H. H. Hendon, and M. C. Wheeler (2006),Large-scale dynamics and MJO forcing of ENSO variability, Geophys.Res. Lett., 33, L16702, doi:10.1029/2006GL026786.

Neelin, J. D., and N. Zeng (2000), A quasi-equilibrium tropical circulationmodel—Formulation, J. Atmos. Sci., 57, 1741–1766.

Randall, D. A., M. Khairoutdinov, A. Arakawa, and W. W. Grabowski(2003), Breaking the cloud-parameterization deadlock, Bull. Am. Meteorol.Soc., 84, 1547–1564.

Randall, D. A., et al. (2007), Climate models and their evaluation, inClimate Change 2007: The Physical Science Basis. Contribution ofWorking Group I to the Fourth Assessment Report of the Intergovern-mental Panel on Climate Change, edited by S. Solomon et al., pp. 589–662, Cambridge Univ. Press, Cambridge, U. K.

Roberts, M. J., et al. (2009), Impact of resolution on the tropical Pacificcirculation in a matrix of coupled models, J. Clim., 22, 2541–2556.

Shaffrey, L. C., et al. (2009), U.K. HiGEM: The new U.K. high-resolutionglobal environment model: Model description and basic evaluation, J. Clim.,22, 1861–1896.

Slingo, J., P. Innes, R. Neale, S. Woolnugh, and G. Y. Yang (2003), Scaleinteractions on diurnal to seasonal timescales and their relevance to modelsystematic errors, Ann. Geophys., 46, 139–155.

Timmermann, A., M. Latif, A. Grotzner, and R. Voss (1999), Modes ofclimate variability as simulated by a coupled general circulation model.Part I: ENSO-like climate variability and its low-frequency modulation,Clim. Dyn., 15, 605–618.

Trenberth, K. E., G. W. Branstator, D. Karoly, A. Kumar, N. C. Lau, andC. Ropelewski (1998), Progress during TOGA in understanding andmodeling global teleconnections associated with tropical sea surfacetemperatures, J. Geophys. Res., 103, 14,291–14,324.

Woolnough, S. J., J. M. Slingo, and B. J. Hoskins (2000), The relationshipbetween convection and sea surface temperature on intraseasonal time-scales, J. Clim., 13, 2086–2104.

Zelinka, M. D., and D. L. Hartmann (2009), Response of humidity andclouds to tropical deep convection, J. Clim., 22, 2389–2404.

�����������������������C. A. DeMott and D. A. Randall, Department of Atmospheric Science,

Colorado State University, Fort Collins, CO, USA.M. Khairoutdinov, School of Marine and Atmospheric Sciences, Stony

Brook University, Stony Brook, NY, USA.J. L. Kinter III, V. Krishnamurthy, C. Stan, and D. M. Straus, Center for

Ocean-Land-Atmosphere Studies, 4041 Powder Mill Road, Suite 302,Calverton, MD 20705, USA. (stan@cola.iges.org)J. Shukla, Department of Atmospheric, Oceanic and Earth Sciences,

George Mason University, Fairfax, VA, USA.

L01702 STAN ET AL.: A GLOBAL CLIMATE SIMULATION L01702

6 of 6