Volume 4, No.4, December 1999 CL/VAR Exchanges

forcings has also been conducted. The principal scien-tific findings from our r(~cent modelling studies relatedto global warming are summarized below.

2. Simulation of the clinlate of the 20th and 21stcenturies

A major recent activity has been the use of theR30 coupled modl~1 to s;tudy climate change over the20th and 21 st centuries. J\ suite of five simulations overthe period 1865-2089 ha1; been completed, in which themodel is forced with estilmates of the observed and pro-jected effective greenhouse gas concentrations and sul-phate aerosols. An c~quivalent CO2 concentration is usedto represent changes in all of the trace greenhouse gases,and changes in aerosol loading are modelled by alteringthe sulface albedo (Mitchell et al., 1995; Haywood etal., 1997). The runs proceed until the late 21 st centurywith equivalent CO2 increasing at the rate of 1 % per yearafter 1990. The ensemble members differ in their initialconditions. which aI.e taken from widely separated pointsin the long control nln. In addition, a suite of six inte-grations is currently underway using new scenarios pro-posed by IPCC-2OCIO (scenarios A2 and 82).

Shown in Fig. I (page 20) is the time series ofobserved global meal' surface temperature (thick, blackline), as well as the time series simulated from the fivemembers of the R30 .~nsemble (various coloured lines).The observed tempc~ratun~ record of the 20th century ischaracterized by an overall warming trend, largely oc-curring in two distinc:t periods (1925-1944, and the late1970s to the present). llie ensemble members largelycapture: the amplitude and timing of the 20th centurywarming. The simulated time series form a spread aroundthe observed record, therl~by offering a perspective onthe role of internal v.mability. Since each model startsfrom independent initial conditions, the internal variabil-ity realized in each of thc~ ensemble members is inde-pendent. Closer examination of the time series for theindividual realizations, however, reveals that one of the

1. Introduction and description of GFDL coupledclimate models

Coupled ocean-atmosphere models have been usedextensively at GFDL over the past several decades for awide variety of climate modelling studies. These includepioneering studies of the transient response of the cli-mate system to increasing greenhouse gas concentrations,as well as studies of paleoclimate and internal variabil-ity of the coupled ocean-atmosphere-land system.

Continuing in this tradition, there are currently twodistinct coupled ocean-atmosphere climate models in useat GFDL for research on global warming and other as-pects of climatic sensitivity and variability. The modelsdiffer in resolution by a factor of approximately 2 butshare similar physics. The R30 coupled model, whichhas been under development for several years, is nowbeing used for the generation of global warming sce-narios and studies of decadal-to-centennial climatic vari-ability. This model has an atmospheric horizontal reso-lution of 3.75° longitude and 2.25° latitude, with 14 lev-els in the vertical. It is coupled to an ocean model withan approximately 2° horizontal resolution, a simple cur-rent-drift sea-ice model, and a "bucket" land model. Auxadjustments are incorporated to reduce climate drift andfacilitate the simulation of a realistic mean state. Theatmospheric component of this model has been studiedextensively. Analyses of a large ensemble of 40-yearexperiments with prescribed observed sea surface tem-peratures (SSTs) show a highly realistic response of themodel to tropical Pacific SST variations, as well as real-istic representations of mid-latitude variability. Selectedoutput from the R30 atmospheric model is available at''http://www.cdc.noaa.gov/gfdl/index.shtmf'. The Rl5coupled model has similar physics but lower spatial reso-lution.

Several long control integrations have recentlybeen performed, exceeding 1000 years for the R30 modeland 12,000 years for the R15 model. A variety of ex-periments with greenhouse gas and sulphate aerosol

CLIVAR Exchanges Volum4~ 4f, No.4, December 1999

cillaltion (AO), are the principal modes of variability ofthe e:xtratropical zonal-mean circulation.) The positivesign of the projecltion is in agreement with observedtrends.

In contrast with t]~e SH, the NH tropospheric cir-culation response involves an equatorward jet shift, anenhanced Aleutian Low, and a negative sea-level pres-sure anomaly over the PlIctic. The response in the Aleu-tian Low and the j\rcti,c SLP agree in sign, but not inmagnitude, with fe:cent observed trends. In particular,the observations show a much larger negative trend inArctic SLP than is seen in the simulations. We are cur-rentl:)' exploring the fa<:tors that underlie this discrep-ancy.

3. Continuing studies v,ith the R15 Coupled Model

The speed of the J.~15 coupled model has alloweda wide range of ret::ent studies with this model. Thesestudi,~s have focused on both radiatively forced climatechani~e and internal variability of the coupled ocean-at-mosphere system. ~;om(~ examples of recent studies arehighlighted belov.'.

a. Large ensemblt~ of climate change simulationsDixon and Lanzante (1999) have recently con-

ducted a nine member ell1semble of greenhouse gas plus

sulphate aerosol experiments covering the period 1765to 2065. The study found a relatively small sensitivityof sin!1ulated surfac(~ air temperature to the year in whichthe model integration started (1765, 1865, or 1915). Thisresull provided the J~atio]!1ale for starting the R30 experi-ment:~ at year 1865. Th(~ ensemble of experiments wasalso used to assess uncertainties in various aspects ofclima,te change, including global mean temperature re-spon~;e and the thermohaline circulation (THC). Themechanisms responsible for the simulated weakening ofthe THC were furthl~r iw..estigated in Dixon et al. (1999)who ~;howed that enhanc:ed atmospheric water flux con-

vergence was the primary factor leading to a reductionof tht: simulated THC. J\dditional analyses are also be-ing conducted on the hydrologic cycle and its responseto grt:enhouse warnling, with emphasis on the near-sur-face c:ontinental clirnate (Wetherald and Manabe, 1999).

ensemble members (Experiment 3) appears to track theobserved time series quite closely, including the observedwarming of the 1920s and 1930s. Since the membersdiffer only in their realizations of internal variability,this suggests that internal variability may have playedan important role in the observed warming of the 1920sand 1930s.

The equilibrium response of this model to a dou-bling of greenhouse gas concentrations is 3.4K, approxi-mately in the middle of the 2.1 K to 4.6K range cited inIPCC (table 6.3, Kattenberg et al., 1995). The agreementbetween the model and observed trends suggests thatthis level of climate sensitivity cannot be excluded based

upon the observational record.The geographical distribution of simulated surface

temperature trends has been compared with the observedtrends for the period 1949-1997 (Knutson et al.,1999).The simulated and observed trends are consistent in mostregions, taking into account the internal variability ofthe trends, as estimated from the model. There are alsoseveral areas which are inconsistent, all of which areregions where substantial cooling has been observed(primarily the midlatitude North Pacific, and parts ofthe Southwest Pacific and the Northwest Atlantic). These

regional inconsistencies are very likely the result of de-ficiencies in one or more of the following: I) the pre-scribed radiative forcing; 2) the simulated response tothis forcing; 3) the simulation of internal climatic vari-

'ability; and 4) the observed temperature record. Distin-guishing between these alternatives is a high priority forfuture research.

The observed temperature trends in this same pe-riod have also been compared to trends generated byinternal climate variability in the control integration. Innearly 50% of the areas analysed (where data wasdeemed adequate for this purpose), the observed warm-

ing trends exceed the 95th percentile of the simulateddistribution of internally generated trends for the samelocation. If the model's simulation of internal climate

variability is accurate, these observed trends are veryunlikely to have occurred due to internal dynamics ofthe climate system.

Studies have also focused on changes in the large-scale extratropical atmospheric circulation under green-house warming. Differing responses were found in thetwo hemispheres. The SH tropospheric response con-sists of a summertime poleward shift of the westerly jet,the storm tracks, and the atmospheric and oceanic mean

meridional overturning. The simulated signal emergesrobustly early in the next century when compared to thecontrol run. The signal-to-noise ratio, however, is rela-tively small because the signal projects strongly and posi-tively onto the model's Antarctic Oscillation (AAO) pat-tern. (The AAO and its NH counterpart, the Arctic Os-

b. Sea ice and climate c,'zangeVinnikov et :il. (1999) compared the simulated

decrease of Arctic sl~a iCi~ in the GFDLRl5 model to theobserved decrease. 'The I:esults (Fig. 2, page 20) demon-strate a remarkable agrc~ement between simulated andobserved trends. Furthc~r, they demonstrated that theobserved decrease in se:l ice extent is outside the rangeof internal variabililty of the model. Additional analysesusing the R30 mod,el provide generally similar results.

Volume 4, No.4, December 1999 CL/VAR Exchanges

Analyses are ongoing to assess Southern Hemispheresea ice changes, as well as changes in snow cover.

c. Multidecadal variabilityDelworth and Mann (2000) have recently com-

pared the simulated multidecadal variability in the NorthAtlantic of the R 15 coupled model to a new multiproxyreconstruction of climate over the last several centuries.In both the model and proxy reconstructions the spec-trum of climate variability has a clear peak on themultidecadal time scale (approximately 60-80 years).The simulated variability involves fluctuations in theNorth Atlantic THC. A comparison of the spatial struc-tures between the model and the observations revealsrelatively good agreement in the North Atlantic sector.Preliminary analyses with the R30 coupled model indi-cate that similar variability exists in the higher resolu-tion model. Such detailed comparisons of model andproxy data offer a promising pathway for increasing ourunderstanding of decadal to centennial scale variability.Delworth and Greatbatch (2000) have also used the Rl5model to demonstrate that such multidecadal variabilityof the THC is partially attributable to stochastic forcingof the ocean through the surface heat flux.

pheric component is nearly identical to the. one employedin the R30 coupled mod,el, to examine the changes intropical climate indillced by the relatively well-docu-mented changes in raljiati've forcing that occurred 21,000years aigo during the last !~lacial maximum. At this time,contiru~ntal ice sheets w,ere greatly expanded, atmos-pheric CO2 was redu(~ed by approximately 25%, and sealevel vIlas more than 100 m lower. When incorporatedinto th,e climate model, these changes produced a meancoolinJ~ of 2 K for the re~:ion from 3005 to 300N.

Comparison of' this simulated cooling with a vari-ety of Ipaleodata indil~ates; that the overall tropical cool-ing is c:omparable to paleoceanographic reconstructionsbased on alkenones and species abundances of plank-tonic rl1icroorganisms, but smaller than the cooling in-ferred from noble gases in aquifers, pollen, snow linedepression, and the isotopic composition of corals. Thepaucit~{ of paleoclimlatic evidence for tropical coolingsmaller than that simlliated by the model suggests that itis unlil,ely that the model exaggerates the actual climatesensitivity in the tro;pics. A more definitive evaluationof the realism of tht: trolpical sensitivity of the modelmust await the resolution of the differences in the mag-nitude of tropical coolin!~ reconstructed from the vari-ous pa1eoclimatic proxie~:.

d. Extreme event in a J 2, 000 year coupled runA control integration of one version of the R 15

coupled model now has been extended beyond 12,000years, thereby offering insights into centennial tomillennial scales of variability. In this extended run ahighly anomalous event occurs in the high latitudes ofthe North Atlantic at approximately model year 3100. Alarge pulse of fresh water, originating in the Arctic, movessouthward through the Fram Strait into the North Atlan-tic. This fresh water pulse is accompanied by reducedoceanic convection and cold surface air temperature (an-nual mean anomalies of -4 K, corresponding to 6 stand-ard deviations below the mean). This event appears tobe an extremely high amplitude realization of a promi-nent mode of internal variability of the coupled system(Del worth et al., 1997). Investigations of the dynamics

of this variability are ongoing.

5. Futulre plans

A major reorganization of the model developmentactivit:r is underway at GFDL. The new modelling sys-tem will consist of both a software infrastructure that isshared across model~; ancl specific models that are con-structed on top of this infr,astructure, including fully cou-pled climate models. 'The latest version of GFDL's Modu-lar Oc~~an Model willI be incorporated into this systemalong Iwith new land and ~;ea-ice models under develop-ment. Two distinct atmospheric dynamical cores are in-cluded in the current realization of this system: a grid-point model on the B..grid, and a standard spectral modelwith the option of grid advection of water vapour andother tracers. A vari(~ty of improvements to the atmos-pheric physics are under active development, includinga new radiative transfer code, boundary layerparaml~terisations, cl~nvective closures and cloud pre-diction schemes with prognostic cloud water. The hopeis that this new structure will reduce the time required todevelop new coupled moldels, ease the transition to newcompulter architectures, provide for more integrated re-search activities witl.1in tihe lab, and foster more exten-sive e;(tramural collaborations. Research on seasonal-to-interannual forec~lsting and on the circulation of themiddl(~ atmosphere, 'which are currently conducted withmodels that are distinct in many different ways from theclimate model, will be using the same modelling system

4. Simulation of the Ice Age -implications for cli-mate sensitivity

A major source of uncertainty in climate modelprojections of future climate involves the sensitivity ofthe climate system to radiative forcing. One way to evalu-ate the realism of a model's climate sensitivity is to simu-late climates of the distant past where sufficient evidenceexists to estimate the changes in climate forcing and re-sponse. In pursuit of this goal, Broccoli (2000) used aatmosphere-mixed layer ocean model, whose atmos-

CL/VAR Exchanges Volume 4, No.4, December 1999

and sharing most physics modules with future climatemodels.

The development of a new climate model for usein long control integrations, global warming scenariogeneration, and paleoclimatic studies is currently focusedon a T42 resolution spectral atmosphere coupled to anocean model with 2° horizontal resolution, except in thetropics where finer resolution is retained to provide abetter ENSO simulation. An ice model with viscous-plas-tic rheology will also be incorporated. A 1 ° version of

the ocean model is also undergoing initial testing. Atpresent, the oceanic, land, and ice components are closerto being finalized than the atmospheric component. Plansalso involve using both grid and spectral atmosphericmodels, at resolutions of TIO6 and higher, coupled tomixed layers or with fixed SSTs (or SSTs generated inlower resolution coupled model global warming scenariostudies) to evaluate how alternative atmospheric physi-cal packages affect climate sensitivity and regional cli-mate change.

References

Haywood, J.M., R.J. Stouffer, R. T. Wetherald, S. Manabe,and '{. Ramaswam:y,1997: Transient response of a cou-

pled model to estimated changes in greenhouse gas andsulfate concentrations. IJeophys. Res. Lett., 24, 1335.

Jone!;, P.D., 1994: :Hemispheric surface air temperaturevariations: a reanallysis and an update to 1993. J. Cli-

mate" 7, 1794-1802.

Kattenberg, A., et a:l. in Climate Change 1995: The Sci-ence of Climate Change (eds. J. Houghton et al.) 285-357 (Cambridge Univ. Press, 1996).

Knutson, T.R., T.L. Delworth, K.W. Dixon, and R.J.Stouffer, 1999: Model assessment of regional surface tem-perature trends (19L~9-97). J. Geophys. Res., in press.

Mitchell, J.F.B., T.I:. Johns, J.M. Gregory, S.F.B. Tett,1995: Climate response to increasing levels of greenhousegases and sulfate ae:rosols. Nature, 376, 501.

Parke:r, D.E.: C.K. JFolland, and M. Jackson, 1995: Ma-rine ~;urface temperature: observed variations and data

requilrements. Clim. Chlmge, 31, 559-600.

Robinson, D. A., 1993: Hemispheric snow cover fromsatellites. Ann. Glaciol., 17,367-371.

Vinnikov, K.Y.,A. R.obock,R.J. Stouffer, J.E. Walsh,C.L.Parki:rlson, D.J. Cavalieri, J.F.B. Mitchell, D. Garrett, andV.F. 2:akharov, 199~1: Detection and attribution of globalwarming using Northern Hemisphere sea ice. Science, in

press.

Wetht~rald,R.E., and S. Manabe, 1999: Detectability ofsummer dryness caused by greenhouse warming. Climatic

Chan,ge, 43, 495-511.

Broccoli, A.J., 2000: Tropical cooling at the last glacialmaximum: An atmosphere-mixed layer ocean modelsimulation. J. Climate, in press.

Chapman, W.L., and J.E. Walsh, 1993: Recent variationsof sea ice and air temperature in high latitudes. Bun. Ame!:Meteo!: Soc., 74, 33-47.

Delworth, T.L., S. Manabe, and R.J. Stouffer, 1997:Multidecadal climate variability in the Greenland Sea andsurrounding regions: a coupled model simulation.Geophys. Res. Lett.. 24, 257-260.

Delworth, T.L., and M.E. Mann, 2000: Observed andsimulated multidecadal variability in the Northern Hemi-sphere. Climate Dynamics, accepted.

Delworth, T.L., and R.E. Greatbatch, 2000: Multidecadalthermohaline circulation variability driven by atmos-pheric surface flux forcing, J. Climate, in press.

Dixon, K.W., and J. Lanzante, 1999: Global mean sur-face air temperature and North Atlantic overturning in acoupled GCM climate change experiment. Geophys. Res.Leu.. 26, 2749-2752.

Dixon, K. W., T.L. Delworth, M. Spelman, and R.J.Stouffer, 1999: The influence of transient surface fluxeson North Atlantic overturning in a suite of coupled GCMclimate change experiments. Geophys. Res. Leu., 26,1885-1888.

EOF -4 8.0%C , ~;:~~:':'

0..-'..1 \, ", ",.-'-' -~ri':~'I" ,,' .,- -.-,.., ."

--

:~-~~:~t 2OE ...5

...-3

OOE

1

1DOE

,1200

3

,...5

,.0.--,.I,.-401 2 0 " ...

Fig. 2: The composite difference of seasonally averaged (a) 850hPa wind (unit ve,ctor==2.5ms-J) and (b) precipita-tion anomalies (mm day-J) for years of above normal all-India rainfall versus years of below normal all-Indiarainfall (see Fig. 1). (c) Thefourth mode of variability extractedfrom an empirical orthogonal function analysis ofseasonal anomalies of 850hPa wind. The magnitude of the wind anomalies is th,~ ,oroduct of the principal compo-nent time series and the components of the wind. Typical variations of this mOG'e a1'e 1-2ms-i. (d) precipitationanomalies (mm day-1) constructedfrom the difference of composites based on years when the principal componenttime series of mode 4 was of above normal versus below normal using +/-0.5 stGrndard' deviations thresholds.

EOF-3 6.6%

"..

200 ...75

.,.

...--15

1000

15

120"

4S

,-75

'00.

R=O.59100

0 ~ -MY... -(+)AlR -(-)AlR ~ \

000 .-L~-40 -2.0 ~O 2.0 40

O~!sol"

UD-

CE~"

+o~

-so

.100Jun 1 Aug1 Sop 1

1981 6-.mu~.u.

Fig. 3: (a) The third mode of variability extractedfrom an empirical orthogonal,fimction analysis of daily anoma-lies of 850hPa wind. Typical variations of this mode are 2-4ms'!. (b ) precipitation ai'lolrUuies (mm day'!) constructedfrom the difference of composites based on days when the principal componenj~ IJf mode 3 was of above normalversus below normal using +/-1 standard deviations thresholds. ( c) Observed daily aU-India rainfall (filled curve,expressed as a percentage of normal) and the principal component time series oJr Jnodt! 3 (red line) for 1987. Bothtime series have been smoothed with a 5-day running mean. (d) Pr.?bability distJ.;,~utionfunctions of the principalcomponent time series of mode 3 for all years (black line), years of above normal all-India rainfall (green line) andbelow normal all-India rainfall (red line). The years of above normal and below normalr all-India rainfall are givenin Fif{. 1.

Jul'

19

Plate 2: Figures from Delworth et al. (see page 15)

2 ".' I I I I I ,--,-.,- I ' ., .-

1.81.61.4

g 1.2~ 1:J"'CO 0.8'-Q)c.. 0.6E~ 0.4

0.2

00.20.4

1865 1890 1915 1940 1965 1990 2~()15 2040

Fig. 1.. Time series of observed (heavy black line) and simulated (various thin colorl~d lil'les) global mean surfacetemperature over the period 1865 to 2040 (1997 for the observations), The simulatt~Q' lines are from 5 independentrealizations of the GFDL R30 coupled model forced with estimates of ob.\'erved greenho'use Irases and sulfate aerosolsuntil 1990, and projections thereafte1: The model output is sampled onlY.for those locat,ions ,and times at which obser-vational data exist. For both model and observations, surface air tempt~rature is used over the continents, while seasurface temperature is used over the oceans, The observed data are a combination OJ!" the Jones (1994) surface airtemperature data and the Parker et al. (1995) sea surface temperature, updated throjj!gh 1997, Anomalies are plottedrelative to the 1880-1920 mean for both the model and observations.

Fig. 2: ObserVed and simulated time series of Northern Hemisphere sea ice extent and sno\v cover (after Vinnikov etal.,1999).

20