Intercomparison and Interpretation of Single …and SCM results using intercomparisons based on well-observed case studies. Corresponding author address: Dr. Ping Zhu, Department of

Intercomparison and Interpretation of Single-Column Model Simulations of aNocturnal Stratocumulus-Topped Marine Boundary Layer

PING ZHU,* CHRISTOPHER S. BRETHERTON,* MARTIN KÖHLER,� ANNING CHENG,# ANDREAS CHLOND,@

QUANZHEN GENG,& PHIL AUSTIN,& JEAN-CHRISTOPHE GOLAZ,** GEERT LENDERINK,��

ADRIAN LOCK,## AND BJORN STEVENS@@

*Department of Atmospheric Sciences, University of Washington, Seattle, Washington�European Centre for Medium-Range Weather Forecasts, Reading, United Kingdom

#NASA Langley Research Center, Hampton, Virginia@Max Planck Institute for Meteorology, Hamburg, Germany

&University of British Columbia, Vancouver, British Columbia, Canada**National Research Council, Naval Research Laboratory, Monterey, California

��Koninklijk Nederlands Meteorologisch Institute, De Bilt, Netherlands##United Kingdom Meteorological Office, Exeter, Devon, United Kingdom

@@Department of Atmospheric and Oceanic Sciences, University of California, Los Angeles, Los Angeles, California

(Manuscript received 15 September 2004, in final form 31 January 2005)

ABSTRACT

Ten single-column models (SCMs) from eight groups are used to simulate a nocturnal nonprecipitatingmarine stratocumulus-topped mixed layer as part of an intercomparison organized by the Global Energyand Water Cycle Experiment Cloud System Study, Working Group 1. The case is idealized from observa-tions from the Dynamics and Chemistry of Marine Stratocumulus II, Research Flight 1. SCM simulationswith operational resolution are supplemented by high-resolution simulations and compared with observa-tions and large-eddy simulations. All participating SCMs are able to maintain a sharp inversion and a mixedcloud-topped layer, although the moisture profiles show a slight gradient in the mixed layer and produceentrainment rates broadly consistent with observations, but the liquid water paths vary by a factor of 10after only 1 h of simulation at both high and operational resolution. Sensitivity tests show insensitivity toactivation of precipitation and shallow convection schemes in most models, as one would observationallyexpect for this case.

1. Introduction

Marine stratus and stratocumulus cloud (MSC),which usually forms from 500 to 1000 m above theocean surface and is a few hundred meters in thickness,plays a crucial role in the global climate system by en-hancing the global albedo and promoting turbulentheat and moisture exchange between the sea surface,the boundary layer, and the overlying troposphere. De-spite improvements in observing, understanding, andmodeling of MSC, serious biases persist in its horizontalextent, vertical structure, and mean albedo simulatedby most general circulation models (GCMs; Gordon et

al. 2000). Likely causes include inaccurate parameter-izations of turbulence and entrainment, cloud fraction,cloud microphysics, and precipitation, as well as insuf-ficient vertical resolution.

As a part of the Global Energy and Water CycleExperiment (GEWEX) Cloud System Study (GCSS;Browning 1993), Working Group 1 (WG1) has orga-nized a series of large-eddy simulation (LES) andsingle-column model (SCM) intercomparison studies ofstratocumulus- and cumulus-cloud-topped boundarylayers. The methodology of GCSS WG1 is to (i) useLES to simulate turbulence–cloud–radiation interac-tions and deduce mean cloud fraction, thickness, andvertical structure in a variety of climatologically impor-tant cloud regimes; (ii) test and improve SCMs usingthe LES results; and (iii) test the fidelity of both LESand SCM results using intercomparisons based on well-observed case studies.

Corresponding author address: Dr. Ping Zhu, Department ofAtmospheric Sciences, University of Washington, Seattle, WA98195.E-mail: [email protected]

SEPTEMBER 2005 Z H U E T A L . 2741

© 2005 American Meteorological Society

MWR2997

The first WG1 intercomparison (Moeng et al. 1996;Bechtold et al. 1996) immediately showed the value ofthis methodology but also some pitfalls in achieving it.This intercomparison focused on an idealized nocturnalnonprecipitating stratocumulus-topped boundary layer(STBL), driven mainly by cloud-top longwave radiativecooling. The goal was to compare the turbulent struc-ture and particularly the entrainment of free atmo-sphere into the boundary layer, as these processes con-trol the evolution of cloud thickness, fraction, and al-bedo. By avoiding solar absorption by clouds andpossible drizzle processes, the intercomparison wassimple and easy to interpret and attracted a broadgroup of LES and SCM modelers. However, only verylimited validation data were available for this idealizedcase, especially regarding the entrainment rate. Onealarming finding of this intercomparison was that therewas a sixfold variation in the simulated entrainmentrate among the participating LES models and a four-fold variation in SCM-predicted entrainment rate. Inmost LES and SCM models, the simulated liquid waterpath (LWP) decreased two- to threefold in the firsthour to unrealistically small values, suggesting exces-sive dry air entrainment. While the scatter in entrain-ment was partly due to the ambiguities in the casespecification of downwelling longwave radiation, thesubsequent WG1 study of an even more idealized andtightly controlled radiatively driven “smoke cloud”boundary layer (Bretherton et al. 1999) confirmed thatthe entrainment rates were overestimated by up to two-fold by LESs and threefold by SCMs compared with alaboratory analog. Very high vertical resolution (5–10m) LESs overestimated entrainment rates less severely.

Recently, Duynkerke et al. (2004) presented an LESand SCM intercomparison study of the diurnal cycle ofmarine stratocumulus by the European Project onCloud Systems (EUROCS). This case was looselybased on observations at San Nicolas Island off theCalifornia coast. Both the mean and diurnal variationof subsidence rate and initial profiles were chosen inpractice to optimize LES–observation agreement. Al-though the model-predicted entrainment rates showless spread than previous studies, LWP varied greatly(more than 10-fold) among participating SCMs. Thediurnal cycle of solar absorption rendered this casemore complex and difficult than the nocturnal STBL.

The Dynamics and Chemistry of Marine Stratocumu-lus (DYCOMS-II) experiment (Stevens et al. 2003a)was specifically designed to accurately measure the en-trainment rates associated with nocturnal marine stra-tocumulus. It took place during July 2001 in the north-east Pacific stratocumulus regime, approximately 500km west-southwest of San Diego, California. ResearchFlight 1 (RF01), documented by Stevens et al. (2003b),provides a particularly attractive intercomparison casefor realizing the vision of GCSS. Figure 1 shows theobserved thermodynamic structure of this case, a well-mixed boundary layer capped by a quite horizontallyhomogeneous cloud layer with negligible drizzle belowcloud base. Estimates of entrainment rates based ondifferent observations and methods were fairly consis-tent, yielding a consensus value of 0.4 � 0.08 cm s�1,which is probably the most reliable in situ estimate evermade for a subtropical STBL. Stevens et al. (2003b)also presented an LES simulation in encouraging agree-ment with the observations. Another intriguing feature

FIG. 1. Observed vertical thermodynamic profiles from RF01 along with profiles used for model initialization (superposed dottedline). Dark lines with dots are the average values over all profiles during the flight. Light horizontal lines indicate the first and thirdquartiles of the observed values.

2742 M O N T H L Y W E A T H E R R E V I E W VOLUME 133

of this case was that the inversion jumps satisfied thebuoyancy reversal criterion for cloud-top entrainmentinstability (CTEI; Randall 1980; Deardorff 1980). Ifpresent, CTEI would lead to a desiccation of the marinelayer. Yet, the observed cloud was quite steady (slightlythickening) throughout RF01. Stevens et al. (2003b) ar-gued that other processes may be simply strong enoughin this case to dominate the tendency of buoyancy re-versal processes to thin the cloud, since CTEI is a com-plicated process that is closely related to boundarylayer turbulent circulations and the details of entrain-ment, rather than just the thermodynamic conditionsacross the interface. Hence, GCSS WG1 selected thiscase to revisit whether LES and SCM simulations ofnocturnal nonprecipitating stratocumulus are nowmore consistent than nine years ago after steady refine-ments in parameterization of boundary layer processes,such as more precise treatments of cloud-top entrain-ment and moist turbulent processes.

This paper reports the results of the SCM part of thisintercomparison study, which includes 10 SCM runs byeight participating groups. The key issues addressedhere are as follows:

• How do SCM simulations compare to LESs when runat comparably high vertical resolution?

• To what extent can SCMs with operational resolutionreproduce the observed STBL structure, entrainmentrate, and LWP under the specified conditions andforcings?

• Do SCMs reproduce the observed lack of drizzle? Do

parameterized precipitation and shallow convectionaffect the simulations?

• What insights are gained from comparing SCM simu-lations with the participating LESs?

Stevens et al. (2005, hereafter STE) present the accom-panying LES intercomparison for this case.

2. Simulation strategy

a. Participating models

Table 1 briefly documents the 10 participating SCMs,including their stratiform cloud and boundary layer tur-bulence parameterizations and three measures of theiroperational vertical resolution: the number of modellevels below 1200 m and in total, and the grid spacing�z at the base of the inversion zi � 840 m. The NavalResearch Laboratory (NRL) SCM did not execute thesimulations with operational resolution; thus no suchinformation was provided. Only the European Centrefor Medium-Range Weather Forecasts (ECMWF), Na-tional Center for Atmospheric Research CommunityAtmospheric Model (NCAR-CAM), Community At-mospheric Model with University of Washington PBLscheme (CAM-UW), and Canadian Centre for ClimateModelling and Analysis–University of British Colum-bia (CCCma-UBC) SCMs use a grid suitable for GCMsimulation; the others have been especially configuredfor this particular simulation, with few levels above1200 m. The ECMWF SCM uses several new physical

TABLE 1. Participating SCMs and their features.

Model Investigator Stratiform cloud scheme PBL turbulence schemeLevels �1200 m/total

�z near 840 m

CCCma-UBC Geng–Austin Prognostic cloud water Bulk nonlocal K-profile 9/35 131 mDiagnostic cloud fraction

METO Lock Diagnostic cloud First order, nonlocal 9/15 175 mExplicit cloud-top entrainment

RPN Lock Diagnostic cloud TKE closure 9/15 175 mKNMI Lenderink Prognostic cloud water TKE closure 19/24 88 m

Diagnostic cloud fractionNCAR-CAM Zhu–Bretherton Prognostic cloud water Bulk nonlocal 6/30 117 m

Diagnostic cloud fractionCAM-UW Zhu–Bretherton Prognostic cloud water TKE closure 6/30 117 m

Diagnostic cloud fraction Explicit cloud-top entrainmentNRL Golaz Diagnostic cloud water Partial third-order closure N/A

Diagnostic cloud fractionNASA-LaRC Cheng Prognostic cloud water Partial third-order closure 12/15 100 m

Diagnostic cloud fractionMPI Chlond Prognostic cloud water TKE closure 12/15 120 m

Diagnostic cloud fraction Explicit cloud-top entrainmentECMWF Köhler Quasi-prognostic total water Eddy diffusivity/mass flux 12/60 201 m

Variance moist thermodynamics Explicit cloud-top entrainment


parameterizations targeted for future introduction intotheir operational forecast model, including a newboundary layer parameterization and a new approachfor predicting cloud fraction and condensate.

Six of the 10 SCMs [CCCma-UBC, Koninklijk Ned-erlands Meteorologisch Institute (KNMI), NationalAeronautics and Space Administration Langley Re-search Center (NASA-LaRC), NCAR-CAM, CAM-UW, and Max Planck Institute (MPI)] determine cloudcondensate prognostically, while the other four calcu-late cloud condensate diagnostically. All models diag-nose subgrid cloud cover fraction based on either rela-tive humidity (e.g., CCCma-UBC, NCAR-CAM) orprobability density function (PDF) of conserved vari-ables such as �l and qt [e.g., Recherche en PrévisionNumérique (RPN), NRL, and NASA-LaRC]. The EC-MWF SCM prognoses the grid-mean and subgrid vari-ance of total water specific humidity, from which cloudfraction and cloud water are diagnosed (Tompkins2002).

The boundary layer turbulence schemes used by theSCMs are quite diverse. They can be classified accord-ing to (i) their formulation of mixing within the STBL,(ii) whether they account for (“moist”) or neglect(“dry”) moist thermodynamics in calculating stratifica-tion of partially or totally saturated layers, and (iii)whether cloud-top entrainment is explicitly computedor an implicit consequence of the model’s mixingscheme.

The U.K. Meteorological Office (METO), CCCma-UBC, and NCAR-CAM SCMs use bulk nonlocal K-profile schemes. The NCAR-CAM and CCCma-UBCscheme is dry and surface-driven, while the METOscheme is moist and accounts for turbulence productionby cloud-top radiative and evaporative cooling (Lock etal. 2000). The remaining SCMs use turbulence closuremodels of various complexity levels. The RPN, KNMI,CAM-UW, and MPI SCMs determine turbulent diffu-sivity based on prognostic turbulent kinetic energy(TKE). The NRL (Golaz et al. 2002; Larson and Golaz2005) and NASA-LaRC (Cheng et al. 2004) SCMs areboth partially prognostic third-order closure schemes.They predict second-order moments of w, �l, qt, and alimited number of third-order moments. NRL predictsonly one third-order moment (w�3), while NASA-LaRC predicts three third-order moments (w�3, ��3

l , andq�3

t ). Both schemes diagnose the other high-order mo-ments based on a PDF determined from the predictedmoments. The new ECMWF model uses an eddy-diffusivity/mass-flux approach, which combines a K-profile diffusion term with a mass-flux term to describenonlocal transports (Tompkins et al. 2004).

Four SCMs (METO, ECMWF, CAM-UW, and MPI)

employ an explicit entrainment parameterization at thetop of buoyancy-driven boundary layers. In the METO,ECMWF, and CAM-UW SCMs, the entrainment pa-rameterization includes both evaporative and radiativecooling effects when the boundary layer is topped byclouds. In the METO and ECMWF models, the en-trainment rate We is explicitly computed followingLock et al. (2000) as

We � A1

V3

�B� A2

�F

�B, 1

where V is a diagnosed turbulence velocity scale, �B isthe buoyancy jump across the inversion, �F is thecloud-top radiative flux divergence, and A1 and A2 areempirical coefficients. The CAM-UW model incorpo-rates �F into the calculation of V, and takes A2 � 0.The MPI entrainment parameterization only considerscloud-top radiative cooling (A1 � 0). Cloud-top en-trainment is not explicitly represented in the KNMISCM but it is implicitly included in the moist turbulencescheme by adjusting the turbulent mixing length (1) via

1l

�1l0

�1

Ce�TKE�N, 2

where N is Brunt–Väisälä frequency, and the constantCe is chosen to obtain reasonable entrainment ratesthrough strong inversions (Lenderink and Holtslag2000).

Models also differ in their operational vertical reso-lution and the organization of the physical parameter-izations. Such a diversity, in a way, makes attribution ofdifferences between model results harder to interpret,but on the other hand, it provides a great opportunity tolook at the behavior of a similar type of parameteriza-tion in a different model framework. In addition to that,intercomparison also allows a close examination of dif-ferent schemes in the same model framework; for ex-ample, the NCAR-CAM and CAM-UW simulationsare particularly illuminating to compare since they onlydiffer in their boundary layer and shallow convectionschemes. A detailed comparison of SCM and globalsimulations between NCAR-CAM and CAM-UW willbe presented in forthcoming papers by P. Zhu and C. S.Bretherton (2005, unpublished manuscript, hereafterZB) and C. S. Bretherton and P. Zhu (2005, unpub-lished manuscript, hereafter BZ).

The companion LES intercomparison consists of 10LES models differing mainly in their numerical algo-rithms and the treatment of subscale motions. A totalof sixteen 4-h simulations were executed under thesame initial conditions and external forcings on a grid


mesh of 96 by 96 points in horizontal with a grid spacingof 35 m and a vertical grid spacing of 5 m or less in thevicinity of cloud top. Statistical analyses were then ap-plied to individual runs to obtain a simulation en-semble. Detailed descriptions of models and analysismethods are provided by STE.

b. Numerical setup

To run an SCM, one needs to specify initial condi-tions, such as dynamic and thermodynamic vertical pro-files, and external forcings, such as horizontal and ver-tical advections. The initial conditions and externalforcings of the DYCOMS-II case for SCM simulationsare specified based on the observations during RF01and are the same as those for the companion LES in-tercomparison. Because of the nature of LES, whichintegrates from a nonturbulent state to a fully turbulentstate, these conditions and forcings have been idealizedin such a way that the important observed features ofSTBL can be best represented by LESs after reachingquasi steady state. STE provides more rationale for theexact case specifications, but these are also includedhere for reference.

Based on observations, the initial thermodynamicprofiles are

�l � � 289 K, if z � zi

297.5 � z � zi1�3 K, otherwise

3

qt � �9.0 g kg�1, if z � zi

1.5 g kg�1, otherwise4

where �l is the liquid water potential temperature; qt,the total specific humidity; z, the height above the sur-face; and zi � 840 m the base of the inversion, equiva-lent to the top of the cloud layer. These profiles com-pare well with the observed vertical thermodynamicstructure shown in Fig. 1, which remained nearly con-stant throughout RF01 (Stevens et al. 2003b).

To obtain the observed boundary layer mean winds(U � 6 m s�1, V � �4.25 m s�1), vertically uniformgeostrophic winds are specified to be Ug � 7 m s�1 andVg � �5.5 m s�1, as are the initial winds. The surfacepressure is taken as 1017.8 hPa.

Since observations showed a fairly uniform large-scale environment with a weak horizontal advection,the horizontal advective tendency of temperature andmoisture are set to zero. The large-scale horizontalwind divergence is set to be D � 3.75 � 10�6 s�1, whichgives a large-scale subsidence rate of about �0.32cm s�1 at the cloud top. This subsidence rate was cho-sen to reconcile the thermal structure above the bound-ary layer with the deduced radiative fluxes and was in a

good agreement with operational analyses for the studylocation and time period.

To minimize the different impact of radiative forcingarising from various radiative schemes, all participatingSCMs and LESs used the following idealized formula-tion for calculating the net longwave radiative flux:

F z � F0e�Qz,� � F1e�Q0,z

� �CpD�z � zi4�3

4� ziz � zi

1�3�, 5

where

Qa, b � ��a

b

�ql dz,

ql is liquid water specific humidity, and and Cp denoteair density and isobaric specific heat of air. Followingthe specification of the LES intercomparison, the SCMboundary layer top zi is defined at the height whereqt � 8 g kg�1. The three terms on the left-hand side ofEq. (3) represent cloud-top radiative cooling, cloud-base radiative warming, and inversion-layer cooling ef-fect, respectively. The parameters �, F0, and F1 are cho-sen as 85 m2 kg�1, 70 W m�2, and 22 W m�2, respec-tively, to match radiative fluxes from a �-four streamradiative code.

Table 2 summarizes the two principal types of simu-lation used for the SCM intercomparison. For both, allmodels were run with their shallow cumulus parameter-ization active. This is because turbulence schemes inseveral SCMs, such as RPN and NRL, include cumulusconvection as a form of turbulence, so that it is notpossible to switch off shallow convection in these mod-els. The only exception was KNMI, which was run with-out its shallow convection scheme for all simulations.

Precipitation and deep convection schemes wereswitched off in models during the simulations. Again,there was one exception. In the ECMWF simulations,the deep convection scheme was turned on, but it wasnot triggered during these simulations; therefore, it ba-sically does not have an effect on simulations. Both

TABLE 2. Characteristics of the numerical experiments. SHF:sensible heat flux; LHF: latent heat flux.

Experiment Integration Surface conditionTime stepresolution

A 6 h SHF � 15 W m�2 5 sLHF � 115 W m�2 10 m

B 48 h SST � 292.5 K 1800 sOperational


experiments A and B were nocturnal (no insolation).Experiment A was designed to compare the SCM simu-lations at high vertical resolution with LES results; thus,it closely follows the LES case specification of a 6-hsimulation with a high vertical resolution of 10 m (anda time step of 5 s, unless otherwise noted), forced by afixed surface sensible heat flux of 15 W m�2 and alatent heat flux of 115 W m�2. Experiment B was de-signed to test SCMs at their operational vertical reso-lution using a 48-h simulation. To reduce model drift, afixed SST of 292.5 K is specified, so that the surfaceheat fluxes are determined interactively in experimentB rather than in experiment A.

The results of experiment B are also compared withan identically forced 48-h simulation using the Univer-sity of California, Los Angeles (UCLA) LES. This LESis further described by STE. For the 48-h simulation, itwas run with no subgrid-scale turbulence scheme ap-plied to scalars and with a vertical grid spacing of 5 mbetween 775 and 1025 m, stretching to 10 m below thislayer and continually stretched above, that is, theUCLA-0 configuration as described in STE. An experi-ment-A simulation with this configuration was quiteconsistent with observations (STE). Unfortunately, theinversion rose through the top of the refined gridaround hour 40, leading to a spurious increase in en-trainment and decrease in liquid water path and cloudcover in the next few hours. Hence, we will showUCLA-LES results only up through hour 36.

The sensitivity of the simulations to modeled precipi-tation processes and shallow convection was also ex-plored by a number of groups.

Because of the limitation of model design, not everymodel can be configured to a specified high verticalresolution or to the requested sensitivity tests. Table 3summarizes the numerical experiments executed byvarious SCMs. The NRL model was targeted for re-gional weather prediction and was not designed to runwith an 1800-s time step. Therefore, it was only used forexperiment A.

3. Simulation results

a. Experiment A

The time variation of the simulated total cloud coverfraction and vertically integrated LWP from experi-ment A are shown in Fig. 2, along with LES-predictedvalues and their standard deviations. STE lumped theLES simulations into LWP quartiles, and found that thehighest-LWP quartile matched the observations muchbetter than the other quartiles. Our measure of the LESsimulation spread was chosen in part because the meanplus one standard deviation of the LWP for the entire

group of LES simulations is a good proxy for the LESupper-quartile mean LWP. In these high-resolutionsimulations, all SCMs except KNMI and NASA-LaRCpredicted 100% cloud cover, in agreement with obser-vations. The KNMI and NASA-LaRC SCMs produceda cloud cover fraction as low as 80% within an hour.The rapid and unrealistic desiccation of the marinelayer in the KNMI and NASA-LaRC simulations sug-gests overefficient entrainment in the presence ofclouds in this case. Several LES models also predictedless than full cloud cover.

By the end of 6 simulation hours, the SCM-simulatedLWPs span a quite large range, from about 12 to over120 g m�2. The large spread in LWP in part reflects thesensitivity of liquid water content to small changes intotal humidity and temperature induced by the differ-ent turbulent transport and microphysics schemes em-ployed in the SCMs. It should be pointed out thatNCAR-CAM and CAM-UW use the same cloud mi-crophysics, and yet produce significant difference inLWPs, indicating that different turbulent transport re-alized in models is a more important factor than micro-physics for causing the large LWP spread in this case.The SCM-mean LWP, however, is comparable to theobservational estimate of 60 g m�2, which is slightlylarger than the LES-mean LWP. Note that the range ofLES-simulated LWPs is comparable to that of SCMs.Hence, as a group, LES models are still not accurateenough to usefully substitute for observations of MSCturbulence, entrainment, and cloud properties.

Figure 3 shows the profiles of �l, qt, U, and V fromexperiment A averaged over the third and fourth simu-lation hours, the same period used for the LES inter-comparison analysis. The �l and qt profiles from thehigh-resolution SCM simulations and LESs agree wellwith each other and with the observed mixed-layer struc-ture. The observed inversion-capped mixed-layer struc-ture shown in Fig. 1 is well simulated by all the SCMs.

TABLE 3. Summary of numerical experiments.

ModelExperiment

AExperiment

BPrecipitation

sensitivity

Shallowconvectionsensitivity

CCCma-UBC Y Y Y YMETO Y Y N YRPN Y Y N NKNMI Y Y Y NNCAR-CAM Y Y Y YCAM-UW Y Y Y YNRL Y N N NNASA-LaRC Y Y Y NMPI N Y Y NECMWF Y Y Y Y


Compared to the LES mean, some SCMs tend tosmooth the sharp jumps a little at the cloud top andhave larger gradients in �l and qt within the mixed layer.While these differences are subtle, they do affect LWPpredicted by these SCMs for a given boundary layermean height and thermodynamic properties. The SCMinversion structure depends on details of its turbulentmixing parameterization; some SCMs require an inver-sion transition layer several grid points deep, while oth-ers (e.g., with explicit entrainment parameterization)can produce inversions only a single grid layer thick.

One useful measure of internal thermodynamic gra-dients across the STBL, also discussed by STE, is themixed-layer moisture difference �qt defined as

�qt � �100m

200m

qt dz � �700m

800m

qt dz.

Figure 4 shows a scatterplot of �qt against LWP forhours 3–4 along with their mean and standard deviationfrom LES and SCM. Except for NCAR-CAM (whichwill be discussed later), large LWP is correlated to small�qt (a particularly well mixed qt profile), presumablybecause this promotes higher qt within the cloud layer.Since all SCMs must pump the same surface moistureflux up through the boundary layer, eddy-diffusivity

reasoning suggests that small �qt (as observed) shouldbe associated with vigorous turbulent mixing across theSTBL.

STE showed that some LES models produced larger�qt in association with marginal decoupling of theboundary layer turbulence into separate cloud and sub-cloud turbulent layers with a cloud-base minimum invertical velocity variance. A similarly simple interpre-tation does not carry over to SCMs. Their diverse tur-bulence parameterizations allow �qt to vary betweenmodels for many reasons, including formulations forturbulent length-scale, turbulence production in partlyor fully cloudy conditions, turbulent transport, and non-local flux-gradient relations.

The horizontal wind profile shown in Fig. 3 evolvessubstantially in the boundary layer from its initial geo-strophic specification and shows some spread acrossSCMs. Both SCM-mean U and V have a noticeable biascompared with the LES-mean wind profiles. This mayreflect inadequate SCM treatment of momentum trans-port, as well as the non–steady state of boundary layerwinds in SCM simulations. The wind profiles from allSCMs are still evolving substantially after 3 h, a timewhen most LES models have reached a quasi steadystate. Note that the plotted mean U and V profiles donot include the CCCma-UBC results, which specified

FIG. 2. Time series of total cloud cover fraction and vertical integrated liquid water pathfrom experiment A. Gray shading indicates the �1 � (standard deviation) range of the LESresults.


constant winds (equal to the geostrophic winds)throughout the simulation. Note that in this experi-ment, there is little potential impact of wind biases onthermodynamic fields since the surface fluxes are speci-fied and the TKE production is dominated by buoyancyrather than shear.

Figure 5 shows the profiles of ql averaged over 3–4simulation hours, which vary substantially betweenmodels to create the LWP variations seen in Fig. 2. LikeLWP, the ql profile is sensitive to the small changes in�l and qt caused by different turbulence schemes. Inmost SCMs, ql increases with height up to a maximumvery near cloud top similar to what is shown in obser-vations (Fig. 1). One exception is the CCCma-UBCsimulation, in which ql reaches its maximum some-where in the upper part of the cloud layer. This is as-sociated with the technique used to calculate the non-local diffusivity, which employs relaxation toward awell-mixed reference state. Several SCMs predict a sub-adiabatic increase of ql with height and a smaller maxi-mum ql than was observed. The NCAR-CAM model

has a secondary ql maximum near 400 m, which appearsto be an artifact of its cloud microphysics scheme. Incontrast, the ECMWF SCM produces a superadiabaticql gradient, generating the observed ql maximum, butonly half as thick as observations.

The simulated STBL evolution is heavily influencedby the boundary layer entrainment rate. For thosemodels in which explicit entrainment parameterizationis employed, the entrainment rate We is readily ob-tained, while for others, it must be inferred indirectly,for example, from the simulated inversion rise rate via

We �dzi

dt� Dzi. 6

The simulated boundary layer top is diagnosed as theheight at which qt � 8 g kg�1 based on a linearly inter-polated qt profile between grid levels. Its evolution inexperiment A is plotted in Fig. 6.

Over simulation hours 2–6, SCMs predicted an inver-sion rise rate dzi/dt in a range of 0.3–0.5 cm s�1, which

FIG. 3. Mean profiles of liquid water potential temperature, specific humidity, and horizontal winds averagedover 3–4 simulation hours from experiment A. Dashed line is the average of SCMs. Solid line is the result from LES(STE). Dotted line delineates the initial state. Light and dark shading indicate the �1 � ranges of inter-LES andinter-SCM profiles, respectively.


is very close to observations (Stevens et al. 2003b), andLESs (STE) near 0.4 cm s�1. These small rise rates inmost SCMs and LESs were consistent with the ob-served rough balance between entrainment and large-

scale subsidence Dzi � 0.32 cm s�1. Compared withobservations and other SCMs, the ECMWF inversionrise rate is notably fast. This is most likely caused by itsentrainment calculation, which is highly sensitive to nu-

FIG. 4. Correlation between moisture difference in the mixed-layer �qt and LWP for hours3–4 from experiment A. Gray shade indicates the range of LES. Filled square box denotesobservations.

FIG. 5. Mean profiles of liquid water specific humidity averaged over 3–4 simulation hoursby SCMs from experiment A. The embedded figure at the right-bottom corner shows theSCM-mean and LES-mean ql profiles and their �1 � ranges (shading) along with observedvalues and initial profiles.


merical discretization around the inversion. This prob-lem could be solved by using the method proposed byLock (2001) and Grenier and Bretherton (2001).

Table 4 lists the resulting hours 2–6 mean entrain-ment rates deduced from (6), as well as the observa-tional and LES estimates. The entrainment rates esti-mated from experiment B (to be discussed shortly) arealso listed in the table. The consistency of We betweenmost models and observations is comparable to thatdocumented in Duynkerke et al. (2004) and is an im-provement from the earlier GCSS study of Bechtold etal. (1996). This consistency in We may result from bothcareful case specification and parameterization im-provements. The latter is evident from the consistencybetween the calculated We from Eq. (6) and the pre-dicted rate from those models, such as METO, CAM-UW, and ECMWF, which employ the explicit entrain-

ment parameterization (not shown here). In contrast tobroadly consistent We, SCM-simulated LWPs show alarge spread at hour 6. This partly reflects the rapidtransient evolution of inversion structure and LWP dur-ing hour 1, when several models undergo up to a four-fold LWP decrease.

It may seem paradoxical that We varies much lessbetween SCMs than LWP, given the central role ofentrainment in the evolution of mixed-layer properties.One perspective considers the feedbacks between theentrainment rate, thermodynamic state of STBL, andcloud properties illustrated by Fig. 7. In each SCMsimulation, the parameterized entrainment process issensitive to cloud liquid water through its impact onturbulence via mixing-induced evaporative cooling andradiative fluxes. This is indicated in Fig. 7 as an arrowfrom LWP to We. Meanwhile, We slowly affects the

FIG. 6. Time evolution of the simulated boundary layer top from experiment A. Shadingindicates the �1 � ranges of LES results.

TABLE 4. SCM-derived entrainment rates and observational and LES estimates. Unit is cm s�1.

NASA-LaRC

CCCma-UBC METO RPN

NCAR-CAM

CAM-UW KNMI NRL MPI ECMWF OBSa LES

Expt A (2–6 h) 0.47 0.27 0.38 0.54 0.32 0.54 0.47 0.42 0.68 0.50 � 0.10b

Expt B (2–6 h) 0.35 0.29 0.29 0.33 0.36 0.37 0.33 0.36 0.28 0.40 � 0.08 0.47c

Expt B (2–36 h) 0.44 0.31 0.39 0.32 0.34 0.35 0.33 0.39 0.47 0.46c

a OBS: Observations.b LES mean.c UCLA-LES.


evolution of the mixed-layer profiles of �l and qt (leftarrow in Fig. 7). Since even a small change in �l and qt

can result in a substantial variation in liquid water con-tent (right arrow in Fig. 7), the LWP rapidly evolvestoward an equilibrium value that sustains roughlyenough entrainment to keep LWP at a nearly steadystate. As discussed by Bretherton and Wyant (1997),this forces the STBL toward approximate moisture andenergy balance between entrainment warming, radia-tive cooling, and surface fluxes. Since the surface fluxesare fixed, as is the radiative cooling as long as the cloudremains optically thick, all SCMs maintain similar en-trainment warming, and hence similar equilibrium We,but with subtly different �l and qt profiles that produceconsiderably different cloud properties. This energybalance and hence the equilibrium We can be changedif the cloud is excessively thinned (as in the KNMImodel) or if there are large numerical errors in treatingentrainment process at cloud top (Lenderink and Holt-slag 2000), as appears to occur in the ECMWF SCM.

Some participating SCMs are capable of maintainingthe persistent clouds observed in this case, while othersare not, although the CTEI criterion is met in all simu-lations. For example, the KNMI and NASA-LaRCmodels simulate partial cloud breakup in response to aparameterized form of CTEI through inversion TKE–We feedback and the negative feedback via radiative forc-ing; that is, thinning clouds correspond to the decreaseof longwave cooling and the reduction of buoyancy,which leads to a further desiccation of the marine layer.

b. Experiment B

The first focus of experiment B is the extent to whichcoarse operational resolution affects the simulation of

bulk cloud layer properties such as the LWP, cloudcover fraction, and mean thermodynamic profiles. Thesecond focus is the long-term evolution of the SCMssubject to constant forcings.

Figure 8 shows the time evolution of SCM-producedtotal cloud cover fraction and integrated LWP fromexperiment B, along with those predicted by UCLA-LES. Even in the first 6 h cloud fractions show a largerspread among SCMs than in experiment A. One mightblame this on the interactive formulation of the surfacefluxes in this experiment. However, the surface fluxescomputed interactively agreed fairly well across SCMs(Fig. 9), though the SCM-predicted surface latent heatfluxes and sensible heat fluxes are generally larger andsmaller, respectively, than those of UCLA-LES, and dodrift over 48 h from the prescribed values of experimentA. Thus, most intermodel variability in cloud propertiesmust be due to the differences between cloud and tur-bulence schemes and in particular their numericalimplementations, aggravated by coarse resolution.

Some SCMs such as RPN, NASA-LaRC, andNCAR-CAM predicted a lower cloud cover fraction inthis experiment. The RPN SCM, an extreme case, pro-duced 100% cloud cover in the high-resolution simula-tion, but less than 40% within a hour of simulation atcoarse resolution, perhaps a consequence of its statisti-cal cloud scheme applied to a vertically underresolvedcloud layer or its vertical cloud overlap assumptions.However, like UCLA-LES, the CCCma-UBC, CAM-UW, MPI, and METO SCMs are still able to produceapproximately 100% cloud cover in the coarse-resolution simulations.

Most SCMs produced 3–6-h average LWPs in experi-ment B, broadly comparable to experiment A (Fig. 10).The most extreme exception is the NCAR-CAM simu-lation in which LWP drops fourfold compared to thehigh-resolution simulation. In section 3c, we show thisresults from the shallow convection scheme employedin the model. The NASA-LaRC and RPN models alsohave rather lower LWP in experiment B, consistentwith their lower cloud cover fraction.

It is worth comparing LWPs in NCAR-CAM andCAM-UW simulations since the difference is solely re-sulted from turbulence and shallow convectionschemes. Figures 2b and 8b show that CAM-UW gen-erated an LWP close to 80 g m�2 in the first 6-h simu-lations in both experiments A and B. Sensitivity testsshown in section 3c indicate that the shallow convectionscheme (ZB; BZ) employed in CAM-UW has little ef-fect on LWP. The insensitiveness of LWP to the modelvertical resolution shown in experiments A and B im-plies the robustness of the CAM-UW moist turbulencescheme (Grenier and Bretherton 2001; BZ; ZB) in rep-

FIG. 7. Schematic illustration of feedbacks among We, LWP,and �l and qt.


resenting MSC cloud fields. In contrast, the NCAR-CAM simulations show a marked sensitivity to themodel vertical resolution. LWPs drop more than sixfoldfrom experiment A to experiment B. Sensitivity testsshown later further indicate that the NCAR-CAM shal-low convection scheme (Hack 1994) plays a significantrole in determining LWP, suggesting that the NCAR-CAM dry turbulence scheme (Holtslag and Boville1993) alone is not sufficient in representing MSC cloudfields although it may predict a similar entrainment rateto that of CAM-UW.

Compared with Fig. 2b, the long-term evolution ofLWPs from different SCMs shows even more variabil-ity, as seen in Fig. 8b. In particular, even though theSCM-mean LWP is similar between experiments A andB, in B the models split after 6 h into one group withconsistently high LWP of 75–140 g m�2 and anotherequally sized group with low LWP of 5–50 g m�2. Thismight be caused by the setup of experiment B, whichleads to long-term drifts of temperature above the in-version due to radiative-subsidence imbalance as evi-denced in temperature profiles shown in Fig. 11. There-fore, SCMs cannot be run to a steady state under thesetup of experiment B. It might be enlightening to for-mulate future SCM intercomparisons designed to focuson the predicted STBL properties in simulations run toa near-equilibrium state.

As in experiment A, the long-term variation in LWPsshown in experiment B (Fig. 8b) is barely correlated tothe mean entrainment rates of hours 2–36 listed inTable 4. For example, in the low LWP group, NASA-LaRC predicted a larger We as one would expect, butRPN and KNMI produced a smaller We. Comparisonsbetween hours 2–6 entrainment rates further indicatethat there is little correlation between the We predictedby a high-resolution simulation and its operational-resolution counterpart. For instance, the ECMWFSCM has the lowest We at operational resolution de-spite having the highest We at high resolution. This cor-roborates the perspective of Fig. 7 that the LWP adjustsin a model- and resolution-dependent fashion to astrong model-independent energy-balance constrainton We.

Figure 11 shows the mean profiles of �l, qt, and ql at3–4 and 35–36 h predicted by various SCMs from ex-periment B compared with those of UCLA-LES. At3–4 h, all SCM simulations and LES still reflect theinitial �l and qt in the mixed layer. The inversion-layerstructure and ql profiles are no longer well resolved bySCMs, and differ between models due to the differentmodel grid levels and the mixing parameterizations.The coarse-resolution models produce a lower peak ofql compared with the high-resolution simulations, butgenerally spread ql over a broader height range to

FIG. 8. Time series of total cloud cover fraction and vertical integrated liquid water pathfrom experiment B.


achieve LWP comparable to the high-resolution simu-lations shown in Fig. 10. Overall, the mixed-layer ther-modynamic structure predicted by the LES is also simu-lated by most SCMs, though with larger vertical qt gra-dients.

Figures 11d–f show that after 35 h, the simulatedprofiles, especially ql, differ more between SCMs. Moresurprisingly, the LES cloud top (1025 m) is actuallyabove the range of the SCM predictions. This implies aslightly stronger entrainment rate in the LES than inthe SCMs, although its LWP was above the SCM me-dian value shown in Fig. 8. This LWP could be sus-tained in the LES despite the dry boundary layer asso-ciated with this stronger entrainment, since the LESgenerated a sharp-topped adiabatic cloud layer and amore precisely well mixed qt profile than most SCMs.

Although all SCMs maintain a fairly well mixedboundary layer below 600 m, there is some spread inthe mean mixed-layer moisture, which may be attrib-uted partially to the different surface moisture fluxesshown in Fig. 9, but mainly to different rates of entrain-ment drying. This is evident from the averaged entrain-ment rates of hours 2–36 listed in Table 4, which havea larger range than those of hours 2–6. For example, theboundary layer deepened much faster and stayed muchdrier in the NASA-LaRC simulation than in the

CCCma-UBC simulation despite the similar surfacefluxes. The LES, with strong entrainment, produces thedriest mixed-layer moisture despite its relatively highsurface moisture flux. However, there is still little cor-relation between the nearly 36-h-averaged entrainmentrate and the hours 35–36 LWP. A more vigorously en-training STBL will have a higher cloud base because itis drier, but also a higher cloud top. Instead, as in thehigh-resolution simulations, each SCM adopts a differ-ent cloud and inversion structure to maintain roughlythe same energy-balancing We. For instance, the RPNsimulation has a slightly lower than average entrain-ment rate, yet the smallest hours 35–36 ql and LWP ofall models. Figures 11d–f suggest that this is associatedwith a “rounded-off” inversion zone that fills two con-tiguous layers, in which cloud could otherwise form,coupled to a turbulence scheme that allows this STBLstructure to entrain vigorously despite the dearth ofclouds.

c. Sensitivity to precipitation and shallowconvection

As marine stratocumulus thickens, drizzle may alsobecome important in the water budgets and turbulentdynamics through latent heating in the cloud layer andsubcloud evaporative cooling. The RF01 intercompari-

FIG. 9. Time variation of the simulated surface latent and sensible heat fluxes fromexperiment B.


son was chosen in part because of the observed lack ofdrizzle, but it is nevertheless possible that in someSCMs drizzle may significantly affect the simulatedLWP and STBL structure. Because of the difficulties inseparating precipitation from other processes in somemodels, only NASA-LaRC, CCCma-UBC, NCAR-CAM, CAM-UW, KNMI, and ECMWF SCMs ex-ecuted the precipitation sensitivity tests at both highand coarse resolutions, while the MPI SCM did only thecoarse resolution sensitivity test. For all SCMs, in ex-periment A, the simulated thermodynamic structure ofSTBL and its associated cloud fields are barely affectedby parameterized precipitation. In experiment B, nosignificant change in cloud cover in response to precipi-tation was simulated by any SCMs, but a weak influ-ence of precipitation on LWP was found in some SCMs.Table 5 lists the LWP with and without parameterizedprecipitation microphysics averaged over 40–48 h fromexperiment B. The CCCma-UBC, NCAR-CAM,CAM-UW, MPI, and ECMWF simulations predictedsmall precipitation fluxes that reduced LWP between6% and 28%, while in the NASA-LaRC and KNMIsimulations, no precipitation was simulated at all.

The weak influence of simulated precipitation onMSC cloud properties shown in this case appears toconflict with other simulations (e.g., Wang et al. 1993)that have found larger effects of suppressing precipita-tion on MSC cloud cover. This is to be expected since

our simulations are based on a nonprecipitating case.An upcoming GCSS WG1 intercomparison will focuson an STBL in which substantial precipitation was ob-served.

We also performed a sensitivity study of the effects ofparameterized shallow cumulus convection on theSTBL simulation. Because of the diverse parameteriza-tions for convective triggering, the shallow cumulusconvection scheme in a SCM may activate in regimeswhere there is no shallow convection in reality. Amongall the participating SCMs, CCCma-UBC, METO,NCAR-CAM, CAM-UW, and ECMWF executed thissensitivity test. In both high and coarse resolution, thesimulations by the METO, CAM-UW, and ECMWFSCMs show little sensitivity to their shallow convectionschemes. The other two models, NCAR-CAM andCCCma-UBC, also show an insignificant influence ofthe shallow convection scheme on the STBL in thehigh-resolution simulations, but a substantial influenceof the convection scheme is found in their coarse-resolution simulations, as shown in Fig. 12. The NCAR-CAM simulations show an enormous sensitivity to itsshallow convection scheme. Without the shallow con-vection scheme, the coarse-resolution NCAR-CAMproduces an unrealistically thick cloud layer almosttouching the ground, indicating that the boundary layerscheme does not generate sufficient turbulence totransport moisture upward efficiently. We speculate

FIG. 10. Comparison of the averaged LWP from 3–6 simulation hours between experimentsA and B.


that the turbulent mixing scheme used in NCAR-CAMtends to produce a colder than realistic cloud layer withlarger LWP and smaller entrainment rate. Since thisbias disappears at high resolution, it might be address-

able by reformulating the NCAR-CAM boundary layerheight diagnosis. In contrast, turning on the shallowconvection scheme increases LWP about 50% in theCCCma-UBC simulations. One possible reason for this

FIG. 11. Mean profiles of liquid water potential temperature �l, total specific humidity qt, and liquid water specifichumidity ql simulated by different SCMs and the UCLA-LES from experiment B. Profiles averaged over (left) 3–4and (right) 35–36 simulation hours, respectively.


increase of LWP suggested by Lenderink and Holtslag(2004) is that the convection scheme tends to produceliquid water through its detrainment process at the topof clouds.

The appropriate coupling between the boundarylayer scheme and shallow convection scheme is not justan issue for correctly simulating the STBL. It becomesmore important in the transition from stratus deck re-gime to trade wind cumulus regime where the interac-tion between stratiform and convective clouds is thekey to determine the thermodynamic structure, cloudcover, and air–sea fluxes in this region.

4. Summary

Accurate simulation of marine boundary layer cloudspresents a long-standing challenge to climate modelingand climate change prediction. This GCSS WG1 inter-comparison study has assessed how well a nonprecipi-tating marine nocturnal stratocumulus-topped mixed-

layer structure can be reproduced by SCMs. Ten SCMswere intercompared using a case based on observationsfrom DYCOMS-II RF01. Although this is an idealizedSTBL case, the initial and forcing conditions are speci-fied in such a way that the important observed featuresof STBL can be best represented by LESs after reach-ing quasi steady state. For this reason, the comparisonbetween SCM simulations and the available observa-tions from field experiments provides an excellent wayto assess the current model’s ability in representingnonprecipitating STBL, and such an assessment shouldnot be affected substantially due to the specific casespecification used in this study. To better separate theeffects of vertical resolution and physical parameteriza-tions, two experiments were executed. In the first ex-periment, a 6-h simulation with all models configuredto 10-m grid resolution was forced by the specified sur-face fluxes and compared with observations and a com-panion study using a suite of similarly forced LES mod-els (STE). In the second experiment, operational reso-

FIG. 12. Time variation of liquid water path and cloud cover fraction simulated by SCMswith shallow convection scheme turned on and off from experiment B.

TABLE 5. Hours 40–48 mean LWP (g m�2) from coarse-resolution simulations.

CCCma-UBC NCAR-CAM CAM-UW NASA-LaRC MPI KNMI ECMWF

Precipitating 89.4 20.1 113.6 28.6 89.3 20.8 93.0Nonprecipitating 113.5 27.7 120.6 28.6 96.4 20.8 115.6


lution and a specified SST were used for a 48-hsimulation. In both experiments the same formulationwas used by all SCMs to calculate radiative cooling, sothat the difference in simulations arising from radiativeforcing was reduced to the minimum.

Given appropriate initial conditions and forcings, allSCMs are able to produce and maintain a mixed STBLcapped by a sharp inversion similar to observations andLESs in simulations with a high vertical resolution com-parable to LES, though the simulated moisture profilesshow a slight gradient in the mixed layer. However,within an hour of simulation, the SCMs disagree mark-edly on their predicted cloud water content and LWP.There is a nearly 10-fold variation in the LWP amongSCMs, although the averaged LWP over all modelsagrees well with observations and LESs.

In operational-resolution simulations, it is not pos-sible for SCMs to simulate the observed sharp inver-sion. As a result, SCMs tend to underpredict cloud wa-ter content. However since they usually also overpre-dict cloud thickness at the same time, such acompensation leads to an integrated LWP (and hencecloud radiative properties) similar to that from high-resolution simulations.

Remarkably, almost all SCMs predicted entrainmentrates within 30% of observational estimates regardlessof resolution, an apparent improvement from SCM in-tercomparisons 5–10 yr ago (e.g., Bechtold et al. 1996).Perhaps this is due partly to more experience with casespecifications, but it is corroborated by recent improve-ments in representing subtropical boundary layerclouds in large-scale models (e.g., Martin et al. 2000).These improvements include a better representation ofa wide variety of turbulence and cloud descriptions andmore careful attention to the interaction of relevantassumptions and their numerical representation. How-ever, our simulations indicate that the improved en-trainment rate estimates do not similarly improve pre-dicted LWP and cloud radiative properties.

A sensitivity test of the inclusion of parameterizedprecipitation processes showed little effect on cloudcover and only a weak influence on LWP in someSCMs. This is as hoped for this essentially nonprecipi-tating case. However, this result does not imply thatcurrent microphysics and turbulence parameterizationare good enough to appropriately represent thicker orless cloud condensation nucleus–rich stratocumulus lay-ers when drizzle is significant. Another sensitivity testindicated that improper triggering of the shallow cumu-lus convection scheme in some models can appreciably(but unrealistically) affect a stratocumulus-toppedmixed layer. Careful design of microphysics and cumu-

lus convection schemes that can function properly to-gether with boundary layer turbulence schemes is acontinuing climate model parameterization issue forwhich SCM simulations and GCSS-style intercompari-sons are powerful tools for progress.

This study suggests that a considerable progress ofrepresenting STBL in large-scale models has beenmade over the last decade owing to the improvementsin turbulence and cloud parameterizations. However,the large continued intermodel variability in LWP inour current intercomparison suggests these improve-ments are still on shaky ground. Models are achievingphysically realistic entrainment rates by rapidly adjust-ing the cloud and inversion structure in model-specificways to bring entrainment warming and drying into arough balance with the combined effects of radiativecooling and surface fluxes. This points to the need forcontinuing careful study of the optimal representationof entrainment and numerical discretization in order torealistically predict LWP and stratocumulus cloud ra-diative properties in GCMs.

Acknowledgments. Zhu and Bretherton acknowledgesupport from NASA Grant NAGS5-10624. Stevens ac-knowledges support from NASA Grant ATM-0097053.This work was performed while Golaz held a NationalResearch Council Research Associateship Award atthe Naval Research Laboratory, Monterey, California.

REFERENCES

Bechtold, P., S. K. Krueger, W. S. Lewellen, E. van Meijgaard,C. H. Moeng, D. A. Randall, A. van Ulden, and S. Wang,1996: Modeling a stratocumulus-topped PBL: Intercompari-son among different one-dimensional codes and with largeeddy simulation. Bull. Amer. Meteor. Soc., 77, 2033–2042.

Bretherton, C. S., and M. C. Wyant, 1997: Moisture transport,lower tropospheric stability, and decoupling of cloud-toppedboundary layers. J. Atmos. Sci., 54, 148–167.

——, and Coauthors, 1999: An intercomparison radiatively-drivenentrainment and turbulence in a smoke cloud, as simulatedby different numerical models. Quart. J. Roy. Meteor. Soc.,125, 391–423.

Browning, K. A., 1993: The GEWEX Cloud System Study(GCSS). Bull. Amer. Meteor., 74, 387–399.

Cheng, A., K.-M. Xu, and J.-C. Golaz, 2004: The liquid wateroscillation in modeling boundary layer cumuli with third-order turbulence closure models. J. Atmos. Sci., 61, 1621–1629.

Deardorff, J. W., 1980: Cloud top entrainment instability. J. At-mos. Sci., 37, 131–147.

Duynkerke, P. J., and Coauthors, 2004: Observations and numeri-cal simulations of the diurnal cycle of the EUROCS stratocu-mulus case. Quart. J. Roy. Meteor. Soc., 130, 3269–3296.


Golaz, J.-C., V. E. Larson, and W. R. Cotton, 2002: A PDF-basedmodel for boundary layer clouds. Part I: Method and modeldescription. J. Atmos. Sci., 59, 3540–3551.

Gordon, C. T., A. Rosati, and R. Gudgel, 2000: Tropical sensitiv-ity of a coupled model to specified ISCCP low clouds. J.Climate, 13, 2239–2260.

Grenier, H., and C. S. Bretherton, 2001: A moist PBL parameter-ization for large-scale models and its application to subtropi-cal cloud-topped marine boundary layers. Mon. Wea. Rev.,129, 357–377.

Hack, J. J., 1994: Parameterization of moist convection in the Na-tional Center for Atmospheric Research community climatemodel (CCM2). J. Geophys. Res., 99, 5551–5568.

Holtslag, A. A. M., and B. A. Boville, 1993: Local versus nonlocalboundary-layer diffusion in a global climate model. J. Cli-mate, 6, 1825–1842.

Larson, V. E., and J.-C. Golaz, 2005: Using probability densityfunctions to derive consistent closure relationships amonghigher-order moments. Mon. Wea. Rev., 133, 1023–1042.

Lenderink, G., and A. A. M. Holtslag, 2000: Evaluation of thekinetic energy approach for modeling turbulent fluxes in stra-tocumulus. Mon. Wea. Rev., 128, 244–258.

——, and ——, 2004: An updated length scale formulation forturbulent mixing in clear and cloudy boundary layers. Quart.J. Roy. Meteor. Soc., 130, 3405–3427.

Lock, A. P., 2001: The numerical representation of entrainment inparameterizations of boundary layer turbulent mixing. Mon.Wea. Rev., 129, 1148–1163.

——, A. R. Brown, M. R. Bush, G. M. Martin, and R. N. B. Smith,2000: A new boundary layer mixing scheme. Part I: Scheme

description and single-column model tests. Mon. Wea. Rev.,128, 3187–3199.

Martin, G. M., M. R. Bush, A. R. Brown, A. P. Lock, and R. N. B.Smith, 2000: A new boundary layer mixing scheme. Part II:Tests in climate and mesoscale models. Mon. Wea. Rev., 128,3200–3217.

Moeng, C.-H., and Coauthors, 1996: Simulation of a stratocumu-lus-topped planetary boundary layer: Intercomparisonamong different numerical codes. Bull. Amer. Meteor. Soc.,77, 261–278.

Randall, D. A., 1980: Conditional instability of the first kind up-side-down. J. Atmos. Sci., 37, 125–130.

Stevens, B., and Coauthors, 2003a: Dynamics and chemistry ofmarine stratocumulus—DYCOMS-II. Bull. Amer. Meteor.Soc., 84, 579–593.

——, and Coauthors, 2003b: On entrainment rates in nocturnalmarine stratocumulus. Quart. J. Roy. Meteor. Soc., 129, 3469–3493.

——, and Coauthors, 2005: Evaluation of large-eddy simulationsvia observations of nocturnal marine stratocumulus. Mon.Wea. Rev., 133, 1443–1462.

Tompkins, A. M., 2002: A prognostic parameterization for thesubgrid-scale variability of water vapor and clouds in large-scale models and its use to diagnose cloud cover. J. Atmos.Sci., 59, 1917–1942.

——, and Coauthors, 2004: Moist physical processes in the IFS:Progress and plan. ECMWF Tech. Memo., 93 pp.

Wang, S., B. A. Albrecht, and P. Minnis, 1993: A regional simu-lation of marine boundary-layer clouds. J. Atmos. Sci., 50,4022–4040.


Documents

Intercomparison and Interpretation of Single …and SCM results using intercomparisons based on well-observed case studies. Corresponding author address: Dr. Ping Zhu, Department of