NASA Technical Reports Server (NTRS) 20140005686: The Aqua-planet Experiment (APE): Response to Changed Meridional SST Profile

1.3 MB·English

by  NASA Technical Reports Server (NTRS)

Journal oftheMeteorologicalSocietyof.Japan,Vol. 91A,pp.57-89,2013 57
DOI:10.2151/jmsj.2013-A03
The Aqua-Planet Experiment (APE): Response to Changed Meridional SST Profile
DavidL.WILLIAMSON
National Ce11terfor Atmospheric Research, Boulder, Colorado, USA
Michael BLACKBURN
Natio11al Ce11trefor AtmosphericScie11ce, U11iversity of Reading, Readi11g, UK
KensukeNAKAJIMA
Faculty ofSciences, Kyushu University,Fukuoka,Japa11
Wataru OHFUCHI
Japan Age11cyfor Mari11e-Earth Sciencea11d Tech11ology, Yokohama, .lapa11
Yoshiyuki0.TAKAHASHI,Yoshi-YukiHAYASHI
Centerfor PlanetaryScience,Kobe,Japan
Facultyof Science, Kobe University,Kobe,Japan
HisashiNAKAMURA
Research Centerfor Advanced Scienceand Technology, The University of Tol.y. o, Tol.y. o, Japan
MasakiISHIWATARI
Graduate School ofScience,Hokkaido University, Sapporo, Japan
John L. McGREGOR
CS/RO MarineandAtmospheric Research,Aspendale,Australia
Hartmut BORTH
TheoreticalMeteorology, UniversityofHamburg, Hamburg, Germany
Volkmar WIRTH
Institutefor AtmosphericPhysics, Universityof Mainz,Mainz, Germany
Conesponding author: Michael Blackbum , National
Centre for Atmospheric Science, Department of Meteo-
rology,UniversityofReading,Earley Gate,POBox243,
Reading,RG66BB,UK
E-mail:[email protected]
©2013,Meteorological SocietyofJapan
58 Journal of the Meteorological Society of Japan Vol. 91A
Helmut FRANK
Research and Development, Deutscher Wetterdienst, Offenbach, Germany
Peter BECHTOLD, Nils P.WEDI
European Centrefor Medium-Range WeatherForecasts, Reading, Berkshire,UK
Hirofumi TOMlTA
Advanced Institutefor Computational Science, RJKEN, Kobe, Japan
Masaki SATOH
Atmosphere and Ocean Research Institute, The University of Tokyo, Kashiwa, Japan
Ming ZHAO, Isaac M. HELD
Geophysical Fluid DynamicsLaboratory, Princeton University,Princeton, New Jersey, USA
MaxJ.SUAREZ
GlobalModeling and Assimilation Office, NASA Goddard Space Flight Center, Greenbelt,Maryland, USA
Myong-InLEE
UlsanNational Institute of Science and Technology, Ulsan, Korea
Masahiro WATANABE, Masahide KIMOTO
Atmosphere and Ocean Research Institute, The University of Tokyo, Kashiwa, Japan
Yimin LIU
State KeyLaboratory of Numerical Modelingfor Atmospheric Sciences and GeophysicalFluid Dynamics,
Institute ofAtmospheric Physics, CAS,Beijing, China
Zaizhi WANG
National ClimateCenter, ChinaMeteorological Administration,Beijing, China
Andrea MOLOD
GlobalModeling and Assimilation Office, NASA Goddard SpaceFlight Center, Greenbelt, Maryland, USA
Kavirajan RAJENDRAN
Centerfor Mathematical Modelling and Computer Simulation, National Aerospace Laboratories, Bangalore, India
Akio KITOH
Meteorological Research Institute, Tr.nkuba, Japan
and
Rachel STRATTON
Met Office, Exeter, UK
(Manuscript received 8August 2011,infinal form 17August 2012)
July 2013 D. L.WILLIAMSON et al. 59
Abstract
This paper explores the sensitivity of Atmospheric General Circulation Model (AGCM) simulations to changes
in the meridional distribution of sea surface temperattu-e (SST). The simulations are for an aqua-planet,a water
covered Eaith with no land, orography or sea-ice and with specified zonally symmetric SST. Simulations from 14
AGCMs developed for Numerical Weather Prediction and climate applications are compai·ed. Four experiments
are performed to study the sensitivity to the meridional SST profile. These profiles range from one in which the
SST gradient continues to the equator to one which is flat approaching the equator, all with the same maximum
SSTat the equator.
The zonal me.ai1 circulation of all models shows strong sensitivity to latitudinal distribution of SST. The Hadley
circulation weakens and shifts polewai·d as the SST profile flattens in the tropics. One question of interest is the
formation of a double versus a single ITCZ.There is a large variation between models of the strength of the ITCZ
and where in the SST experiment sequence they transition from a single to double ITCZ. The SST profiles are
defined such that as the equatorial SST gradient flattens, the maximum gradient increases and moves poleward.
This leads to a weakening of the mid-latitude jet accompanied by a poleward shift of the jet core. Also consid-
ered ai·e tropical wave activity and tropical precipitation frequency distributions. The details of each vary greatly
betwe.en models,both with a given SSTand in the response to the change in SST.
One additional experiment is included to examine the sensitivity to an off-equatorial SST maximum. The up-
ward branch of the Hadley circulation follows the SST maximum off the equator. The models that form a single
precipitation maximum when themaximum SSTisonthe equator shifttheprecipitation maximum offequatorand
keep it centered over the SST maximum. Those that form a double with minimumon the equatorial maximum
SST shift the double structure off the equator, keeping the minimum over the maximum SST. In both situations
only modest changes appearin the shifted profile of zonal average precipitation. When the upwai·d branch of
the Hadley circulation moves into the hemisphe.re with SST maximum, the zonal average zonal, meridional and
vertical winds all indicate that the Hadley cell in the other hemisphere dominates.
Keywords comparison of atmospheric general circulation models (GCMs); sea sw·facetemperature profile;Hadley
circulation;intertropical convergence zone (ITCZ);precipitation
and of the experimental configurations to which they
I. Introduction
are applied. The hierarchy has two distinct roles: an
The Aqua-Planet Experiment (APE) consists of evaluation role in the development and testing of
a series of Atmospheric General Circulation Model complex atmospheric models, and a conceptual role
(AGCM) simulations on a variety of water covered in linking complex models with theory and observa-
planets which otherwise have Earthlike physical char- tion. This context of APE is discussed in more detail
acteristics. The experiment involves a comparison of by Blackburn and Hoskins (2013), so only a summary
a number of AGCMs which are used for or are being isgiven here.
developed for climate and numerical weather predic- In the conceptual context, theory and more ideal-
tion applications. The AGCMs are complete as devel- ised models provide constraints on the character of the
oped for Earth applications but are applied to an ideal- global circulation expected in APE. The zonally aver-
ized water covered planet with no land, orography aged SST distributions in APE are broadly similar to
or sea-ice. The Sea Surface Temperatures (SST) are Earth , so the equator-to-pole temperature difference is
specified to have simple distributions and are fixed in expected to give rise to a jet stream and storm track
time. A modest number of SST fields, to be described in mid-latitudes . Theory suggests that the location and
below, are specified in order to study the response of strength of the stormtrack will closely follow low-level
aqua-planet simulations to changes in the SST fields. baroclinicity associated with the mid-latitude SST
In this paper we consider the response to changes inthe gradient. Recent idealised modelling studies (Lu et al.
meridional SST profile and the response to shifting the 2010; Butler et al. 201 l) have highlighted the impor-
maximum SSToff the equator. tance of changes in static stabi lity in influencing the
APE was conceived by Neale and Hoskins (2000a) latitude of maximum baroclinic i nstability, suggesting
as one component of a modelling hierarchy of that increased sub-tropical static stability may be one
increasing complexity, both of the models themselves mechanism that causes the storm tracks to migrate
60
Journalofthe Meteorological Society ofJapan Vol. 91A
polewards in simulations of the 21st centmy climate et al., provided motivation for Nea le and Hoski ns
warming(Yin 2005; Lorenz and DeWeaver 2007). (2000a) to propose a benchmark test suite of AGCM
In the tropics, the analytic model of Held and Hou aqua-planet experiments. In this evaluation context
( 1 980) predicts that a Hadley circulation with equa- of the modelli ng hierarchy, APE is a bridge between
torial ascent must exist when the latitudinal profile of experi ments with models of reduced complexity that
temperature in equilibrium with the underlying SSTs are used in the development of individual model
is steeper than quartic, in order to maintain a themal components, and realistic simulations of Earth's
wind balance consistent with angular momentum climate using complete AGCMs, coordinated through
conservation. The tropical SST profiles in APE were the Atmospheric Model lntercomparison Project
designed to include this limiting case: they vary from (AMlP, described in its original form by Gates 1992).
a profile denoted PEAKED in which the mid-lati- APE therefore provides a test-bed for the interaction of
tude SST gradient extends to the equator, through a dynamics and physical parameterizations in AGCMs
CONTROL which is quadratic and QOBS which is which is simpler than Earth yet contains the complete
closer to the observed profile in the Pacific,to a quartic sub-gridscale parameterization package developed for
profile denoted FLAT. The tropical circulation is there- Earth applications. For this reason, APE is sponsored
fore expected to consist of a Hadley circulation with by the Working Group on Numerical Experimenta-
equatorial ascent for the steeper SST profiles, but this tion (WGNE) which is jointly sponsored by the World
should break down in the limiting FLATcase to give a Meteorological Organization (WMO) Commission
state closer to radiative-convective equilibrium at each on Atmospheric Science (CAS) and World Cl imate
latitude. Research Program (WCRP). The goals of APE are to
Held and Hou ( l980) considered only forcing and expose inter-model differences, to stimulate research
SSTs symmetric about the equator in their analytic to understand their causes and to provide a benchmark
and idealised numerical model study. Later, Lindzen of the behaviors ofthe current generation of models.
and Hou (l988) found that moving the SST maximum This paper considers the simulated responses to
only a few degrees off the equator resulted in a marked varyingthelatitudinalprofileofSST,usingexperiments
asymmetry of the Hadley cells, with rising on the based onthefivezonally symmetric SSTdistributions
summer side of the equator and a much strengthened defined by Neale and Hoskins. A companion paper,
cell crossing the equator into the winter hemisphere . Blackbum etal.(20l3),discussestheCONTROL SST
In order to test this asymmetry in AGCM simulations, experimentinmoredetail.
Neale and Hoskins (2000a) proposed a fifth zonally
2. Experimental design and participating models
symmetric forcing case in which the SST maximum is
moved 5°off the equator, CONTROL SN. The latitudinal stmctures of the five zonally
Previous modelling studies using aqua-planet symmetric SSTfieldsareshown inFig. l.Thesimple
configurations have confirmed that the tropical circula- formulae definingthese distributionsarenotrepeated
tion and the location of the Inter-Tropical Convergence here but were originally given in Neale and Hoskins
Zone (ITCZ) produced by AGCMs do depend on SST. (2000a) and are repeated in Blackburn and Hoskins
The most systematic study, by Hess et al. (l993), used (2013) and Williamson et al. (2012).The four latitu-
a number of widely varying tropical SST profiles and dinally symmetric distributions, labeled PEAKED,
two different convective parameterizations in a single CONTROL, QOBS andFLAT,haveamaximum SST
model. Hess et al. and other studies found that the trop- of 27C at the equator and differ in their meridional
ical circulation and ITCZ location also differs between gradients approaching the equator. All experiments
models, with some models producing a double-ITCZ specifyaSSTofOCpoleward of60°latitude,withno
straddling the equator even when SST is peaked on sea ice.The PEAKED experiment has a strong SST
the equator. Such a double-ITCZ was found in the gradient continuing to the equator while the FLAT
first modem aqua-planet simulation, by Hayashi and experimenthasitsmaximum gradientpolewardof20°
Sumi (1986).An unreal istic double-ITCZ is also found latitudeandisquarticin latitude,givingabroadequa-
in many simulations using coupled atmosphere-ocean torialSSTmaximumandflatterprofileapproachingthe
GCMs (e.g., Lin 2007). As discussed in Blackburn and equator. The CONTROL SSTis quadratic in latitude
Hoskins (2013), the origin of this phenomenon and its and QOBS is the average of CONTROL and FLAT.
relationship to uncoupled simulations with prescribed Based on theirexperience with theMet Officemodel
SSTsare not well understood. (HadAM3,Popeetal.2000)appliedtotheseSSTfields,
A number ofthese studies, inparticular that ofHess ealeand Hoskins(2000a)designatedtheCO TROL
July 2013 D. L.WILLIAMSON et al. 61
AMlP II (Wangetal. l99S; Liangand Wang 1996).
Thesimulations arerun for3.S years foreach exper-
iment. The analysis is made over the last 3 years, omit-
25
ting the first 6 months as spi n-up. A model-simulated
state, from either a previous aqua-planet integration
20 or an earth -like si mulation, should be used for the
...-... initial conditions. In independent tests of the spin-up
(_)
in several models, aqua-planet climate equilibrated in
r15
en a matter of weeks from these types of initial condition,
en so a 6 month spin-up is considered adequate. everthe -
10 -PEAKED less, the mode IIing groups were instructed to check that
-CONTROL equili bration was achieved during the discarded initial 6-
-QOBS month period. Daily time-series of global averages were
5 -FLAT requested for a number of variables, which allows a
--CONTROL_5N
gross check that equilibrium was reached. A varietyof
0 data sets from thefinal three years of the simula- tions
60 40 20 0 -20 -40 -60 was requested for the APE data archive. This l ist is
LATITUDE considerably longer than that suggested by Neale and
Hoskins (2000a) and is given on the APE web site
(http://cl imate.ncas .ac.uk/ape/) and summarized in the
Fig. 1. Zonal average SST for PEAKED,
ATLAS.
CONTROL,QOBS,FLATandCONTROLSN,°C
Simulations from 16 models were contributed to the
APE data base. These models are listed in Table l . The
as the standard experiment. However, QOBS is closest table includes the commonl y accepted model name, the
to the observed zonal mean boreal winter SST in the group that contributed the data and the location of the
Pacific.Thefifth SST, denoted CONTROLSN, distorts group's home institution. The group symbol is used to
the CONTROL SST by shifting the maximum to S0N identify the models in figures, tables and discussions.
while holding the transition to OC fixed at ±60° lati- Not all models contri buted simulations for all SST
tude. As discussed in the introduction, the four hemi- distributions . The last two col umns of Table l indicate
spherically symmetric SSTs are intended to study the those that contributed PEAKED, CONTROL, QOBS
issue of the intensity of the lTCZ and the formation and FLAT and those that contribut ed CONTROL and
of a single ITCZ on the equator versus a double ITCZ CONTROL SN. Those are the models analyzed here.
spanning the equator. The hemispherically asymmetric The models include established models that have been
SST is intended to examine the response of the ITCZ used in production applications such as numerical
when the maximum SSTis noton the equator. weather prediction and IPCC simulations, and some
The APE design included certain model specifica- newer, more novel models that have been less well
tions in order to obtain uniformity across the models. tested in real-world applications. Table 2 summarizes
Complete details are available on the APE web site the dynamical cores, the methods used for water vapor
(http://climate .ncas .ac.uk/ape/) and in an atlas of model transport and resolutions of the participating models .
simulation statistics (Williamson et al. 2012), hereafter Table 3 lists major choices in the model parameter-
referred to simply as the ATLAS. Many aspects of the izations of convection and boundary layer turbulence .
design are also briefly summarized in Blackburn et al. Brief descriptions of the models with references for
(2013) and Blackburn and Hoskins (2013). Values for their algorithms are available in the ATLAS.
basic geophysical constants and parameters are recom- The simulations from the CONTROL experiment
mended . The insolation is perpetual equinoctial and are presented in a companion paper (Blackburn et al.
symmetric about the equator, but includes the diurnal 2013). Most of the figures i ncluded there and in the
cycle. Values are specified for the solar constant and ctment paper are extracted from the ATLAS. Because
C0 concentration . Recommendations are given for of the number of experimen t s considered (five) and
2
other well mixed gasses as well as for the mass of the number of models involved (fourteen of the sixteen
dry atmosphere. Ozone is specified as a zonally and submitted) we can only provide a sampling of the
meridionally symmetric latitude-height distribution properties of the aqua-planet simulations and an indi-
dete1mined from the annual mean climatology used in cation ofthewide varietyofmodel behaviors. Detailed
62
Journalofthe Meteorological Society ofJapan Vol. 91A
Table l. Participatingmodels
Group Location Mode l Tuned PCQF3 C5N4
symbol TOA?1
x x
AGU Japan (consortimn) AFES No
x x
2 CGAM Reading,UK HadAM3 Yes
x x
3 CSIRO Aspendale,Australia CCAM No
x x
4 DWD Offenbach/Mainz, Germany GME Yes
x x
5 ECM-CY29 Reading, UK IFS cy29r2 No
x
6 ECM-CY32 Reading,UK IFS cy32r3 No
7 FRCGC Yokohama,Japan NICAM No
8 GFDL Princeton,USA AM2.l Yes x x
9 GSFC Maryland,USA NSfPP- 1 No x x
x x
JO KIJAPAN Japan (collaboration) CCSR/NIES5.7 Yes
x x
II LASG Beijing,China SAMIL No
12 MIT Cambridge,USA MIT-GCM No x x
x x
13 MRI Tokyo,Japan MRI/JMA98 No
x x
14 NCAR Boulder,USA CCSM-CAM3 Yes
15 UKMO (N48) Exeter,UK pre-HadGAMI Weakly2 x
16 UKMO (N96) Exeter,UK pre-HadGAMI Weakly2 x
1i.e.wasthe topof atmosphere radiative balanceoptimised forpresent dayfauth climate?
2Dming thedevelopment phase ofHadGEMl theTOAfluxes ofAMIP runs weremonitored to check that they did
not widelydivergefrom balancebut werenot activelytuned.
3Submitted results fi'om PEAKED,CONTROL,QOBS and FLATexperiments.
4Submitted results from CONTROLandCONTROL 5N experiment.
Table2. Dynamicalpropertiesofparticipatingmodels
Group Dymunical Water vapor Horizontal Vertical
symbol core transport resolution resolution
AGU Eulerianspectral Eulerian spectral T39 L48
CGAM lat-Ion grid point Euleriangrid 3.75°x 2.5° L30
CSIRO C-C1 semi-Lag2 semi-Lag -210km(C48) LIS
DWD icosahedral grid semi-Lag grid -10 L31
ECM-CY29 semi-Lagspectral semi-Lag grid T159 L60
ECM-CY32 semi-Lag spectral semi-Lag grid T159 L60
FRCGC icosahedral Eulerian Eulerian -?km L54
GFDL lat-Ionfinitevolume finitevolume 2.5°x 2° L24
GSFC lat-Ion gridpoint Eulerian centered 3.75° x 3° L34
KlJAPAN Eulerian spectral semi-Lag grid T42 L20
LASG Eulerian spectral Eulerian grid R42 L9
MIT cubedsphere Euleriangrid -280km L40
MRI Eulerianspectral Eulerian spectral T42 L30
NCAR Eulerianspectral semi-Lag grid T42 L26
UKMO(N48) semi-Lag lat-Ion grid semi-Lag 3.75°x 2.5° L38
UKMO(N96) semi-Laglat-Ion grid semi-Lag 1.875°x 1.25° L38
1C-C denotes conformal cubic
2semi-Lag denotes semi-Lagrangian
July 2013 D. L.WILLIAMSON et al. 63
Table3. Parameterizationsofparticipatingmodels
Group PBL Shallow Deep
symbol convection convection
AGU Mellor-Ymnada None Ermumel
CGAM Smith Gregory-Rowntree Gregory-Rowntree
CSIRO Holtslag-Boville None McGregor
DWD Louis Tiedtke Tiedtke
ECM-CY29 Louis-Beljaars Tiedtke Bechtoldetal.2004
ECM-CY32 Louis-Beljaars Bechtoldetal.2008 Bechtoldetal.2008
FRCGC Mellor-Ymnada None None
GFDL Lock RAS! RAS
GSFC Louis RAS RAS
KJJAPAN Mellor-Yamada None Pan-Randall
LASG Local vert diffusion None Manabe
MIT Mellor-Ymm1da RAS RAS
MRI Mellor-Yamada Randall-Pan Randall-Pan
NCAR Holtslag-Boville Hack Zhang-Mcfarlane
UKMO(N48) Lock/Richardson Gregory 1990/Grant Gregory 1999
UKMO(N96) Lock/Richardson Gregory 1990/Grant Gregory 1999
1RAS denotes relaxed Arakawa-Schubert.
anal yses are left for individual studies of specific summarize the properties of the current set ofAGCMs.
aspects. We summarize the collective APE model In this set, no onemodel isagood match forthe multi-
behavior for many statistics with a multi-model mean . model mean, and in general the individual models all
The statistics are always calculated on each model 's differ from the multi-model mean.
submitted data grid, then linearly interpolated to a 1° To indicate variations among the APE models we
latitude-longitude grid to calculate the multi-model present the inter-model standard deviation . The stan-
mean . A more complete description of the process is dard deviation of the statistics is also calculated after
given in the ATLAS. The models are weighted equally the statisti cs for each model are calculated on the
for the mean even though with some statistics there are submitted model grid and interpolated to the l 0 grid.
model outliers. The sample size is large enough that the As with themulti-model mean,the models areweighted
impact of outliers is generally small, although there are equally for the standard deviation even though with
a few exceptions. One subtle aspect is that the models some statistics there are model outliers. These outliers
included in the multi-model mean for the CONTROL do have a noticeable effect on some of the standard
depend on which experiments it is being compared to. deviations . The corresponding plots for all individual
Thus in Blackburn et al. (2013), which considered only models are available in the ATLAS but we do not
the CONTROL experiment, the multi-model mean provide specific figure numbers here. When discussing
includes all models in Table 1 except FRCGC which is the multimodel mean and inter-model standard devi-
an extremely high resolution model and was only run ation, we often describe in words the source of the
for 30 days. The multi-model mean for the PEAKED , variability between models. This is our subjective
CONTROL, QOBS, FLAT comparison here includes evaluation of the plots from all the individual models
all models with an X in the next-to-last column i n contained in theATLAS. These overviews are intended
Table 1. The multi-model mean for the CONTROL SN, to emphasize the great variation between the models in
CONTROL comparison later includes all models with most ofthe statistics examined.
an X in the last column i n Table l . Thus plots of the As noted above and as can be seen in the ATLAS,
CONTROL for the multi-model mean will not be iden- there are three or four models that are relative outliers
tical in the different comparisons. The multi-model in manymetrics.These do affectthe multi-model mean
mean should not be thought of as a best estimate of the climate to some extent, but probably have a larger
aqua-planet climate. It is simply a convenient way to effect on the inter-model standard deviations. Never-
64
Journalofthe Meteorological Society ofJapan Vol. 91A
theless we perform all analyses with the complete set of energy flux and the atmospheric component balances
models. All contri buted models were accepted i nto the the difference between TOA and surface net fluxes.
APE data base with no overall quality criterion. In fact, Globally averaged net fluxes have been removed and
applying a quality cut-off would be particularly diffi- thetwo hemispheres have been averaged.
cul t si nce the APE climate is unknown. Perhaps criteria We first compare the poleward energy transport for
could be based on the quality of Earth-like simulations the PEAKED, CONTROL, QOBS and FLAT experi-
but even then only egregi ous outl iers could be safel y ments. Thetransports ofthe CONTROL were discussed
discarded. A PE outl iers only become apparent in the in Blackbum et al. (2013) and compared to estimates
comparative analyses, and there is a remote possi bility of the annual average transport for Earth from Fasullo
that they provide a more accurate indication of the and Trenberth (2008). Figure 2 shows the multi-model
APE cl imate. We provide plots of some statistics for mean zonal-time average poleward energy transports
individual models as wel l. indi vidual model plots are for the four experiments (solid lines) along with the
made on each model's submitted grid, i.e. they show Fasullo and Trenberth (2008) estimates (dashed lines).
the statistics before interpolation to the I 0 grid. All Pl us to minus one inter-model standard deviation is
statistics for all models are available in the ATLAS. shaded. For each experiment, the atmospheric transport
Neale and Hoskins (2000b) considered the response shows little variation between the APE models which
ofa singlemodel tothese different SSTfields. Here we are strongly affected by the fixed SST. This implies
determine how robust their results are by comparing that the large range in the TOA net energy flux (not
14models. shown here, but available in the ATLAS and reflected
in the total transport) passes through the atmosphere
3.f Sensitivity to meridional SSTprofile
andaffects the implied ocean transport.
In this section we consider the changes in the simu- The TOA net energy flux is relatively insensitive
lations as the meridional profile of SST changes from to SST. ln the sequence from PEAKED to FLAT, the
the PEAKED case, in which the mid-latitude gradient net downward flux increases slightly in the tropics and
continues to the equator,to the FLATcase, in which the decreases in the extra-tropics. The decrease extends
equatorial maximum is very broad and SST is quartic beyond the latitude of zero net flux near 40° which
in latitude, i.e. from PEAKED to CONTROLto QOBS therefore shifts slightly equatorward. This leads to
to FLAT. In this sequence, the region of the primary a small equatorward shift in the pattern of the total
SST gradient steepens, narrows, and moves poleward. transport going from PEAKED to FLAT seen in Fig.
For these four cases the SST is symmetric about the 2, mainly associated with an increase in tropical trans-
equator and the radiative forcing is symmetric about port. The equatorward shift brings the latitudinal
the equator.Thus the time average zonal means arealso distribution of the total transport closer to that of Earth
symmetric about the equator.Therefore we average the although the overall amplitude increases at the same
two hemispheres together to reduce the noise slightly time so that the maximum total transport in the FLAT
and plot one hemisphere only, in the sense of the case is alittle over20%larger than that of Earth.
Northern Hemisphere . In addition, the abscissa adopts TheTOAnetenergyfluxand total poleward transport
sineof latitude in order to provide detail in the tropical show almost no variation through the SST sequence
regionswhich exhibit more structure inmanystatistics. poleward of 60° latitude where in all cases the SST
fields are flat and equal to 0°C. In this region APE
3.1 Poleward energy transport atmospheric and implied oceanic transports in Fig. 2
Blackburn et al. (2013) compared the insolation for also show little variation through the SST sequence.
APE to the annual average insolation for Earth. Since The total transport is stronger than that of Earth, ulti-
the SST is fixed, the fraction of insolation reaching the mately driven by the stronger gradient of insolation,
surface is lost. Nevertheless , an implied ocean pole- but the atmospheric transport is less than that of Earth
ward heat transport can be computed to balance the and the oceanic is larger. This is consistent with the
net surface energy flux if the global average energy APE lower boundary being open ocean with SST set
imbalance is neglected . For each model the total pole- to 0°C, which results in the APE atmosphere gaining
ward transport by the atmosphere plus an implicit more energy from the underlying surface than does
underlying ocean has been computed to balance the net Earth's atmosphere.
top of atmosphere (TOA) energy flux. These are aver- ln mid-latitudes, changes in the atmosphere-ocean
aged to give the APE multi-model mean transport. The transport partition are broadly consistent with changes
implied oceanic component balances the net surface intheunderlyingSSTgradient, shown inFig. l. In the
July 2013 D. L.WILLIAMSON et al. 65
PEAKED CONTROL
10 10
-TOTAL APE TOTAL -APE
ATMOSPHERE -- EARTH ATMOSPHERE --EARTH
8 -OCEAN 8 OCEAN
6 6
$:
.,, .,,
4 4
0 0
..- ..-
2 2
0
-2 -2
80 60 40 20 0 80 60 40 20 0
LATITUDE LATITUDE
QOBS FLAT
10 10
-TOTAL -APE TOTAL -APE
ATMOSPHERE -- EARTH ATMOSPHERE --EARTH
8 -OCEAN 8 OCEAN
6 6
$:
.,, .,,
4 4
-0 -0
..- ..-
2 2
0
-2..........._.. .___.__ . .... . _, -2............ .. .___.__ . _. _ . _,
80 60 40 20 0 80 60 40 20 0
LATITUDE LATITUDE
Fig. 2. Multi-model mean poleward energy transport for the PEAKED, CONTROL, QOBS and FLAT experi-
ments (solid lines). Plus to minus one inter-model standard deviation is shaded. Total transport is that required
to balancethetop of atmosphere net radiative flux; ocean transport is that implied to balance the surface net
flux; atmospheric transportis thedifference between the two. Annual mean observational estimates for Earth
are included for comparison , from Fasullo and Trenberth (2008) (dashed). Va lues shown are averaged over the
Nmthern and Southern hemispheres.
PEAKED case, the atmosph ere transport i s weak and CONTROL and FLAT the changes are smaller. The
the ocean transport strong compared to the other cases, atm osphere trans port m oves slightly po leward and
consistent withthe weaker mid-latitude SST gradient, strengthens, al ong with the max imum SST gradient,
since the fixed temperature difference between mai nl y by strengthening the po l eward flank of t he
the equator and 60° is linear in latitude. Between transport peak.
66
Journalofthe Meteorological Society ofJapan Vol. 91A
The most dramatic response of the atmo- sphere- those of Ear1h, with the atmosphere component being
ocean transport partition to SST occurs in the tropics. remarkablyclose.
The total transport increases only slightly goi ng
from PEAKED to FLAT, but a broader tropical SST 3.2 Zonal mean state
maximum produces weaker transport in the atmo- ln this section we discuss how the multi-model
sphere and stronger transport in the ocean equatorward mean simulated state changes as the meridional SST
of 30°. The QOBS case is closest to Earth, particu- profile changes for the four zonally and hemispheri-
larly for the atmosphere transport, with the total (and cally symmetric experi ments. Blackbum et al. (2013)
ocean) transport being slightly larger than observed. compared the CONTROL experiment with the Ea11h's
The change in transport partition with SST is so climate. We first consider the time average, zonal
strong that, in the deep tropics for the PEAKED case, average precipitation forthe individual models and for
the atmosphere transports more energy than the total the multi-model mean. This provides an overview of
with equatorward transport in the ocean, while for the the inter-model variability. We then consider the time
FLATcase the entire transport is carried by the ocean. average, zonal average of the three-dimensional atmo-
The absence of poleward atmospheric transport in the spherebyexaminingthemulti-model mean. Finallywe
deep tropics in the FLAT case is a first indication that describe thevariation between themodels and summa-
the Hadley circulation does indeed break down in the rize it by showing the intermodel standard deviation
APE models for this SST profile, as predicted by the of the zonal-time averages. The multi -model mean
theoretical model of Held and Hou (1980), leaving the includes all models except FRCGC and UKMO(N 96)
atmosphere close to radiative-convective equilibrium asindicated inTable I.
at each latitude in thedeeptropics.
The modest sensitivity of total poleward transport in a. Precipitation
the tropics to SST is due to a large degree of compen- The zonal average, time average prec1p1tation
sation between changes in net shortwave and longwave providesagoodoverviewoftheresponseoftheHadley
fluxes at the TOA in the tropics (shown in the ATLAS). circulation to changes in the SST profile. Figure 3
Incontrast, changes in net shortwave flux at the surface shows the zonal average total preci pitation for the
dominate over those in longwave and are augmented multi-model mean and for the individual models for
by changes in the turbulent heat fluxes, dominated by the PEAKED, CONTROL, QOBS and FLAT experi-
the latent heat flux . These contributions lead to the ments. The top two rows focus on the tropics, 0° to
large opposing changes in separate atmosphere and 25° latitude, where most of the response in precipita-
implied ocean poleward transports.As the tropical SST tionoccurs.Thebottom tworowsshowthesub-tropics
maximum broadens going from PEAKED to FLAT, the andextra-tropics, 15°to90°latitude,with anenlarged
net solar flux at the TOA and surface, which is down- preci pitaton scale. To make the curves of the indi-
ward, increases in the equatorial region and decreases vidualmodelsmorevisibletheyaredividedacrosstwo
in the subtropics, due to migration of deep cloud from panels(e.g.,firstandsecondrows)foreachexperiment
the equator to the subtropics. The maximum surface (columns).Themulti-model mean (overallmodels)is
latent heat flux similarly migrates poleward from the includedonbothpanelsofthepair.
deep tropics in PEAKED to beyond 20° in FLAT. Considering first the multi-model mean, the tropical
The reduction in near equatorial latent heat flux in the preci pitation maximum becomes weaker and broader
FLAT case is associated with weaker surface winds, from PEAKED to QOBS. It remains a single peak
again indicative of a breakdown of the large scale on the equator through the first three experiments,
overturning circulation. The weaker evaporation more although it is basically flat within 5° latitude of the
closely balances local precipitation (shown later), also equator in QOBS. Inthe FLAT casethepeak moves off
indicating approximate radiative-convective equilib- the equator to around 14° latitude giving a wide double
rium in the deep tropics. The maximum latent heat flux structure with symmetric peaks about the equator.
is influenced by both wind strength and the latitudinal The PEAKED SSTdistribution yields a single ITCZ
gradient of saturated specific humidity, q*( Ts), so in all models,but themaximum precipitation rate varies
remains on the poleward flank of the subtropical trade greatly between models, by almost a factor of three.
winds. The large downward net surface flux at the The FLAT case yields a pair of precipitation maxima
equator in the FLAT case leads to the strong implied symmetric about the equator in all models except
ocean transport. LASG which retains a single but very weak maximum
Overall, all the transports in QOBS are most like ontheequator. In most modelsthe off-equator maxima