Max-Planck-Institutfür MeteorologieREPORT No. 93Observed and simulated zonal mean precipitation over l andDJF precip250 d -------------------Obs T21 T4220015010050II60o -r 2-90 30 0 30Lat i tude60 90preci mrn/moJJA precip250 -< ------------------Obs ----- T21 ----- T42E 200 -E 150 -100 -50 --30-60 0Lat i tude30-90 60 90SIMULATION OF THE PRESENT-DAY CLIMATEWITH THE ECHAM MODEL:IMPACT OF MODEL PHYSICS AND RESOLUTIONbyE. ROECKNER • K. ARPE • L BENGTSSON • S . BRINKOPL DOMENIL • M . ESCH • E. KIRK • F. LUNKEIT • M. PONATERB. ROCKEL • R. SAUSEN • U . SCHLESE • S . SCHUBERT • M. WINDELBANDHAMBURG, OCTOBER 1992AUTHORS:Erich RoecknerKlaus ArpeLennart BengtssonLydia DümenilMonika EschMax-Planck— Inst i tutfü r MeteorologieEdilbert KirkFrank LunkeitSilke SchubertMartin WindelbandUniversität HamburgMeteorologisches Inst i tutBundesstraße 55D-2000 Hamburg 13FRGUlrich Schlese Deutsches Kl imarechenzentrum (DKRZ)Bundesstraße 55D-2000 Hamburg 13FRGSabine BrinkopMichael PonaterRobert SausenDeutsche Forschungsanstaltfü r Luft— und Raumfahr t (DLR)Institut f ü r Physik de r AtmosphäreD-8031 OberpfaffenhofenFRGBurkhardt Rockel GKSS Forschungszentrum GeesthachtInstitut fü r PhysikPostfach 1 1 60D— 2054 GeesthachtFRGMAX-PLANCK-INSTITUTFÜR METEOROLOGIEBUNDESSTRASSE 55D— 2000 HAMBURG 13F.R. GERMANYTel.: +49 (40) 4 11 73-0Telex: 211092mpime dTelemail: MPI. METEOROLOGYTelefax: +49 (40) 4 11 73-29BSimulation of the present-day climate with the ECH AM model:Impact of model physics and resolutionE.Roeckner 1 , K.Arpe 1 , L.Bengtsson 1 , S.Brinkop 4 , L.Dümenil 1 , M.Esch’. E.Kirk 2 , F.Lunkeit 2 ,M.Ponater4 , B.Rockel , R.Sausen 4 , U.Schlese 3 , S. Schubert 2 and M. Windelband 21 Max-Planck-Institut für Meteorologie, HamburgMeteorologisches Institut der Universität, Hamburg3 Deutsches Klimarechenzentum (DKRZ), Hamburg4 Deutsche Forschungsanstalt für Luft- und Raumfahrt (DLR), Institut für Physik der Atmosphäre,Oberpfaffenhofen5 Institut für Physik am GKSS Forschungszentrum GeesthachtAbstractA detailed description of the third-generation atmospheric general circulation model ECHAM is presen-ted. The climatology of ECHAM3, as simulated at low and high resolution (T21 and T42, respectively),is compared with ECMWF analyses (1981-1990) and also with the climatology of two earlier low reso-lution versions ECHAM 1 and ECHAM2.At low resolution, the impact of the improved model physics is evident primarily in the simulated time-mean state which is more successfully reproduced by the more recent model versions, particularly byECHAM3. The impact of increased horizontal resolution can be identified not on] v by a generally impro-ved time-mean circulation, particularly in the Southern Hemisphere, but mosbsignificantly by the incre-ased level of high-frequency variability. In the low-frequency range, however, the impact of increasedhorizontal resolution is modest, and all models fail to reproduce the observed level of low-frequencyintraseasonal variability.ISSN 0937 -10601. INTRODUCTIONGeneral circulation models of the atmosphere and ocean are designed to derive the climate from firstphysical principles such as the conservation laws of angular momentum, energy and mass. The solutionof the governing equations can, however, only be obtained by approximate methods which require adiscretization of the space-time continuum. In addition to the truncation of the continuous spectrum, thebasic error sources are generated by inaccuracies of the numerical solution method and. furthermore, bythe need to consider the statistical effect of the unresolved scales on the resolved scales of the model,generally referred to as the parameterization problem.During the past decades, considerable success has been achieved to improve both, numerical weatherprediction (NWP) models and general circulation models (GCMs), by increasing the spatial resolution(parallel to the increase of computer power), by improving the numerical solution techniques, and byrefining the parameterization schemes.There is no doubt that increasing the spatial resolution reduces the forecast errors of NWP models (Sim-mons et al., 1989). On the other hand, there is no clear evidence that the climate simulated by GCMs willsignificantly improve with increasing resolution (Boer and Lazare, 1988; Boer et al., 1991; Boville,1991). On the basis of a model intercomparison study which was undertaken to document the skill ofGCMs to reproduce the modem climate, Boer et al. (1991) concluded that the models are more sensitiveto changes of the parameterization schemes than to changes of the horizontal resolution.The objectives of the present study are similar to those of Boer et al. (1991 ). however, instead of com-paring a diversity of models with different resolution, different numerics and different physical parame-terizations, we use only one model which allows us to isolate the relative importance of the processesunder consideration in a more systematic manner. Furthermore, we will not only compare the simulatedand observed time mean climates but also the respective intraseasonal fluctuations.The model used in this study is the so-called ECHAM model (Roeckner et al., 1989) which evolved fromthe numerical weather prediction model developed at the European Centre for Medium Range WeatherForecasts (ECMWF). An old version of the ECMWF model (the so-called cycle 17 version, operationalin 1985, in the following referred to as “ECHAMO”) has been extensively used for climate studies at theMax-Planck-Institute for Meteorology and the University of Hamburg (Fischer, 1987, v.Storch, 1988;Fischer, 1989; Barnett et al., 1989; Latif et al., 1990; Metz and Lu, 1990; Ponater et al., 1990; Robertsonand Metz, 1990; Barnett et al., 1991). In these studies, a number of model weaknesses have been identi-fied as, for example, a cold bias in the tropical troposphere of more than 5K and a warm bias over thepolar regions. The reduction of the meridional temperature gradient caused a weakening of the meanzonal flow. In particular, the near-surface flow on the Southern Hemisphere was severely underestimated.The intraseasonal variability was significantly smaller than observed, particularly in the low frequencyrange, but also at higher frequencies in the storm strack region of the North Atlantic, and the circulation2over Antarctica showed a pathological behaviour. Moreover, part of the model physics was constrainedto present-day climate conditions, such as the transfer of heat and water within the soil or the paramete-rization of CO2 absorption, so that climate change experiments could not be conducted.In the second generation model (ECHAM1 and ECHAM2), several parameterization schemes were eit-her replaced (radiation, cloud formation, land surface processes, horizontal diffusion) or additionallyimplemented (gravity wave drag in ECHAM1), and minor changes were introduced in the numerics (for-mulation of the pressure gradient term) and in the vertical transport schemes (convection, turbulent dif-fusion). The ECHAM1 model, coupled to an oceanic GCM, has been predominantly used to study theclimatic impact of increasing levels of greenhouse gases in the atmosphere (Cubasch et al., 1992) or sootemission from burning oil wells in Kuwait (Bakan et al., 1991). The ECHAM2 model differs fromECHAM1 basically with respect to a weaker orographic forcing (no gravity wave drag and use of a grid-mean orography instead of an envelope orography)The third generation model (ECHAM3) which was released in 1991, employs a comprehensive massflux scheme for cumulus convection (Tiedtke, 1989), instead of the Kuo scheme in the former models,and the effect of clouds on the turbulent mixing in the boundary layer is taken into account.All of these models use a comparatively high vertical resolution (19 layers, except for ECHAMO whichhas only 16). The horizontal resolution is typical of GCMs that have been used so far for long term cli-mate simulations, namely triangular truncation at zonal wavenumber 21 (T21 ), corresponding to a Gaus-sian transform grid of approximately 5.6°. The more recent model version (ECHAM3) has beendesigned primarily for higher resolution such as T42, for example.All of these models have been integrated in the so-called uncoupled control mode by specifying the seasurface temperatures (SSTs) and sea ice limits according to the observed climatological seasonal cycleso that interannual fluctuations of SST and sea ice have been artificially suppressed. The simulated timeis 10-20 years, and the model climates are compared with ECMWF analyses for the time period of 1981-1990.The sections 2 and 3 provide a fairly detailed description of the dynamics and physics used in ECHAM3.For more information and for technical details, we refer to the model documentation (DKRZ-Report No.6). Section 4 contains a list of the various model versions that have been used for performing long termcontrol experiments. The sensitivity of the ECHAM model to changes of physical parameterizations andto doubling the horizontal resolution is discussed in section 5, and a summary (section 6) concludes thisreport.32. MODEL DYNAMICS2.] GOVERNING EQUATIONS AND SOLUTION METHODLike the ECMWF numerical weather prediction model, the climate model ECHAM is based on the pri-mitive equations. Prognostic variables are the vorticity £ and the divergence D, the temperature T, thespecific humidity q v and the logarithm of the surface pressure In p s . In addition to the prognosticvariables of the ECMWF model, the cloud water mixing ratio q w (liquid and ice) is also a prognosticvariable in ECHAM. Basic equations are the vorticity and the divergence equations,= rb7 (/7v+Pv) ’ ä ( + )+ ’ (2 ’ U)= + + + <2.1.2)U \L )wherein Rj7\.<2.1.3)and3v ~= (2,1.4)The remaining equations are the hydrostatic approximation,9 T T1 + ” “ 1 Mv Pk \ c P d ) )the continuity equation,1Jt nPs = -p fr <2L7)5 0the equation for the water vapour mixing ratio3<7v = _ U dq? _V Q v _a ( l -n 2 j 3X a dp. n 3r| + ?. +4and finally the equation for the cloud water mixing ratio, q w - q e + q t , where q c is the liquid fractionand q t is the solid fraction,U_ w_V w_ . w3r ~ a ( l -nV X ’■ (2.1.9)In these equations t is the time, a the radius ot the earth, X the longitude and p = sin
n*)L n = T (2.2.3)0 for (n 16 are damped.The diffusion coefficient kx varies for different variables and levels. While the lower levels (6 to 19)use the same kx , the upper levels are stronger damped,£x ( / ) = kx - f ac ( l ) (2.2.4)where I denotes the vertical level. For fac = 1 andT21 resolution (ECHAM2, ECHAM3) the dampingtime 1 / ( k x • L n ) of the shortest scale is 1.12 days for the vorticity, 0.22 days for the divergence and5.59 days for temperature, humidity and cloud water, respectively. For T42 resolution the respectivetimes are 0.30, 0.06 and 0.76 days. Table 2 shows the amplification factor for the six uppermost levelsand Figure 3 shows the damping times of the vorticity due to the horizontal diffusion as a function of thetotal wave number n and level I .Table 2: Amplification factor of the horizontal diffusionI 1 2 3 4 5 6fac 16 16 8 4 2 1A modified diffusion scheme is used for the temperature to avoid unrealistic warming of mountain topsand excessive summer precipitation associated with substantial mixing in the vicinity of steep mountainslopes. Only the deviation of the temperature field from the ICAO standard atmosphere is inserted intothe diffusion procedure.101 - - -2 —3 —4 —5 —6 —7 —8 —9 —10 —11 —12 —13 —14 —15 —16 —17 —18 —19 —QJ>wavenumber2 4 6 6 10 12 14 16 18 20 22 24 26 28 30 32 34 36 3B 40 42— 1-1> 3 =4 —5 -6 —7 —8 —9 —10 -11 —12 —13 -14 —15 —16 —17 —IB —19 —123456789101112131415161718195 ,I o' 0.10.221000 1001 • I * I • I ‘ I ’ I ' I ' I 1 I ' I 1 I 1 I ’ I 1 I14 16 10 20 22 24 26 28 30 32 34 36 38 40 42wavenumberFigure 3 Damping times for the vorticity due to horizontal diffusion for TH (upper panel) andT42 (lower panel) horizontal resolution. Contours for 0.1, 0.2, 0.5, 1, 2, 5. 10, 100, 1000 days.11The diffusion coefficients were chosen such that the slope of the spectral kinetic energy is close to obser-vations (see Figure 4). Moreover, as a result of the new diffusion scheme together with a time step of 40(24) minutes for T21 (T42) resolution, the need for additional damping to avoid violation of the CFL-cri terion is considerably reduced as compared to the ECMWF model.long t ime mean j an kinet ic energy 500 hPa10' T-r-raioio'•5a'• V?io10ioV.I10i o40n20------------------- T21- ....................... T42------------------- observationFigure 4 Mean kinetic energy spectra at 500 hPa during January for observations (ECMWF,mean of 1980 - 1989), for ECHAM2/T21 (mean of a 20 year control run with prescribedclimatological sea surface temperature) and for ECHAM3/T42 (mean of a 10 year simulationwith the observed sea surface temperatures of the years 1979 - 1988).123. MODEL PHYSICSThe model physics is formulated in a simplified parametric form, either because the respective processcannot be resolved explicitly (e.g., turbulent transfer, cumulus convection or cloud microphysics), orbecause a more accurate treatment would exceed the computer resources presently available (e.g., thecomputation of radiative transfer).The radiation scheme (section 3.1) uses a broad-band formulation of the radiative transfer equations withsix spectral intervals in the terrestrial infrared and four intervals in the solar part of the spectrum (Henseet al., 1982; Rockel et al., 1991). Gaseous absorption due to water vapour, carbon dioxide and ozone istaken into account as well as scattering and absorption due to aerosols and clouds. The cloud optical pro-perties are parameterized in terms of the cloud water content which is a prognostic variable of the model.The vertical turbulent transfer of momentum, heat, water vapour and cloud water (section 3.2) is basedupon the Monin-Obukhov similarity theory for the surface layer and the eddy diffusivity approach abovethe surface layer (Louis, 1979). The drag and heat transfer coefficients depend on roughness length andRichardson number, and the eddy diffusion coefficients depend on wind shear, mixing length andRichardson number which has been reformulated in terms of cloud-conservative variables (Brinkop,1991, 1992)The effect of orographically excited gravity waves on the momentum budget (section 3.3) is paramete-rized on the basis of linear theory and dimensional considerations (Palmer et al., 1986; Miller et al.,1989). The vertical structure of the momentum flux induced by gravity waves in stably stratified flowover irregular terrain is calculated from a local Richardson number which describes the onset of turbu-lence due to convective instability and the breakdown approaching a critical level.The cumulus convection scheme (section 3.4) comprises the effect of deep, shallow and mid-level con-vection on the budget of heat, water vapour and momentum (Tiedtke, 1989). Cumulus clouds are repre-sented by a bulk model including the effect of entrainment and detrainment on the updraft and downdraftconvective mass fluxes. Mixing due to stratocumulus convection is parameterized as a vertical diffusionprocess (Tiedtke et al., 1988) with eddy diffusion coefficients depending on the cloud water content,cloud fraction and relative humidity jump at cloud top.The prediction of stratiform clouds (section 3.5) is based upon the cloud water equation (cf., section 2.1)including sources and sinks due to condensation/evaporation and precipitation formation by coalescenceof cloud droplets and sedimentation of ice crystals (Sundqvist, 1978; Roeckner et al., 1991). Sub-gridscale condensation and cloud formation is taken into- account by specifying appropriate thresholds ofrelative humidity depending on height and static stability.The soil model (section 3.6) comprises the budgets of heat and water in the soil, the snow pack over landand the heat budget of permanent land ice and sea ice (Dümenil and Todini, 1992). The heat transferequation is solved in a five-layer model assuming vanishing heat flux at the bottom. Vegetation effectssuch as interception of rain and snow in the canopy and stomatai control of evapotranspiration are para-13meterized in a highly idealized way (Blondin, 1989). The runoff scheme is based on catchment conside-rations and takes into account sub-grid scale variations of field capacity over inhomogeneous terrain.3.1 RADIATIONThe radiative transfer model is based on a two-stream approximation described by Kerschgens et al.(1978) and Zdunkowski et al. (1980). From the radiative transfer equations for the solar part of thespectrum, eqs. (3.1.1) and (3.1.2), and for the terrestrial part, eqs. (3.1.3) and (3.1.4), Hense et al. (1982)derived a broad-band formulation with spectral intervals listed in Table 3.= [ l -w( l -ß ) ] -©ß - -ß o äs o e 8/g| ' (3.1.1)= röß - - [ l - rö ( l -ß ) ] =- + ( l - ß 0 )ö>S u e 8 /g ” (3.1.2)= 1 (M+ -B )d£> |1(3.1.3)4m~ = 1 -B )J8 p (3.1.4)where M + and M~ are the upward and downward flux densities, 8 is the optical depth, co is the singlescattering albedo, ß and ß 0 are the backscattering parameters for diffuse and direct radiation respectiv-ely, 1 / p is the diffusivity factor (2 for solar and 1.66 for terrestrial radiation), S o is the solar irradiance,p 0 is the cosine of the solar zenith angle, and B is the Planck function.Table 3 Spectral intervals and absorbers of the radiative transfer modelTerrestrial SolarSpectral Interval(pm) Absorber Spectral Interval(pm) Absorber3.96 - 7.98 H 2 O 0.215 - 0.685 O 37.98 - 8.89 H 2 O, dimer 0.685 - 0.891 h 2 o8.89-10.15 Oj, dimer 0.891 - 1.273 h 2 o10.15-11.76 H 2 O, dimer 1.273 - 3.580 h 2 o, co 211.76-20.10 H 2 O,CO 220.10- 100. h2 oThe parameterization of the optical parameters in a cloud-free atmosphere is taken from Hense et al.(1982) for the solar spectrum and from Eickerling (1989) for the terrestrial spectrum. In the terrestrialpart scattering is neglected.14Terrestrial radiationThe optical depth 8 is defined as- n8 = exp flyClogwJ 7 (3.1.5)L; = 0with the effective absorber mass u i of the 1 th component,a t a rn p i oWl = p . ( ) ( ) d (3.1.6)Pq 1where a , a T are coefficients derived by inverting the results of an "exact" reference radiationmodel, p . , p and T denote concentration of the 1 th component, pressure and temperature, respectively,averaged over an atmospheric layer with geometrical thickness d , p Q and T() are reference values setto 1013 hPa and 273.15 0 K, respectively.With these definitions and assuming that the Planck function is a linear function of 8 , the radiative trans-fer equations can be solved as described by Hense et al. (1982).Solar radiationFor the transfer of solar radiation, absorption (subscript a) and multiple scattering (subscript 5 ) are takeninto account. The relevant parameters are the optical thickness 8 , the single scattering albedo to and thebacksc altering coefficients ß , ß Q . The backscattering coefficients ß and ß 0 are derived from theory(Kerschgens et al., 1978). Scattering and absorption coefficients of stratospheric, urban and maritimeaerosols are taken from a data set provided by Shettle and Fenn (1975). Multiplication by d gives theappropriate optical thicknesses $“ eroso1 , tf* eroso1The optical thickness d8a as for gas absorption is a function of the effective absorber amount and is definedsimilarly to the terrestrial radiation (see above). The optical thickness 8 8as for Rayleigh scattering of gasmolecules is given by&s as = N ■ A R ■ d (3.1.7)where N is the mean number density and A R is the scattering cross-section of molecules.For diffuse radiation, the optical thickness 8 is calculated from&“f = + + + (3.1.8)where a a , b a , a s and b s are coefficients derived by inverting results of an "exact" reference model.The single scattering albedo is defined as15d) = 1 - -77 (3.1.9)8* zFor direct radiation, the optical thickness S d l r at a level k is calculated by adding the optical thicknessesabove this levelkS“ r = £ (5 f s + 8“ crolo ' + 8f s + 5* eTOS<’') (3110)j = 1where 6 8a a s depends on the zenith angle 0 . & ga as is parameterized for 0 = 35 0 and 0 = 75 0 only. Opticalthicknesses for other zenith angles are calculated by linear interpolation.Cloud optical propertiesThe emissivity e of clouds in the terrestrial region is described by(3.1.11)e = 1-e L pwhere k = 0.084 m g’ 1 is the mass absorption coefficient (Stephens, 1978), d is the geometrical cloudthickness and m is the water or ice content in gm~3 .Optical thickness o , single scattering albedo co, and back-scattering coefficient ß are the optical para-meters considered in transfer of solar radiation. For both cloud phases, the parameterization of Stephens(1978) is adopted:0.2633 + 1.7095 • In [log (m • d) ] (X < 0.685 pm)log (5 ) = { (3.1.12)1 0.3492 + 1.6518 - In [log (m ■ d) ] otherwiseSingle scattering albedo and back-scattering coefficients are taken from Kerschgens et al. (1978).Surface albedoOver snow-free land areas an annual mean background albedo is specified from satellite data (Geleynand Preuss, 1983). According to more recent analyses (e.g., Dorman and Sellers, 1989) the satellite-based estimates are too low in the tropics and at middle latitudes (North America, Eurasia). This bias hasbeen removed by applying an empirical correction in those areas where the satellite data suggests valuesof less than 15 %. The minimum values thus obtained are around 12 % for tropical rain forests.In snow-covered areas the surface albedo is modified according to»S na Surf = a Sb + ( a s~ a slJ ’ Sn + Sn* (3.1.13)where a s is the snow albedo (see below), a Sb is the background albedo, Sn is the simulated snow depth(in water equivalent) and Sn* = 0.01 m is a critical snow depth. For Sn » Sn* the surface albedo16approaches the albedo of snow.The albedo of snow and ice surfaces ( a ) is a function of surface type (t s ), surface temperature (T v )and fractional forest area (fly) over land. For T s >T m = 273.15 K (i.e., for melting of snow or ice), a sis fixed at a relatively small value, a s = & Smin ( ts , af), whereas a s is larger, a s. = Smax (t$, a f> * f° rcold surfaces (T s = a f a S m in a f= + " a f> ' a Smi„ = 0 > 1a SmaA a f) = a f a S ma a f = 1 ) + U ~ a f> ' a Sm a/ a f=In the temperature range T Q < T s < T m , a s = a s (T s , t s , af is obtained by linear interpolationT s -T 0a S “ a Smax a Smax a Smin ' 7 _T (3.1.15)m 0with and a Smin as specified in Table 4 according to the estimate of Robock (1980) for sea ice andsnow (without the solar zenith angle correction, however) and according to Kukla and Robinson (1980)for land ice.Over sea, the albedo is constant (0.065) for diffuse radiation but depends on the solar zenith angle 0 forthe direct beam0.05fy = ---------------50 cos0 + O.15(3.1.16)with the limitation a 50 <0.15.Table 4 Minimum and maximum surface albedos (Robock, 1980; Kukla and Robinson, 1980)referring to the temperature ranges T s > T m = 273.15 K and T s for (3.2.2.b)where p T is the pressure at the top of the boundary layer and y h = (u, v) is the horizontal windvector.The definition of the drag coefficient C x depends on the height z above the ground at which y h and Xare taken (the natural choice is the model's lowest level) and on the stability of the layer. X s representsa value of X at the surface.X may be identified with each of the five variables u, v, q v , q w and 5 (dry static energy), y is defined as:s = C pd ( \+ (C pv /C pd - \ )q v ) -T+g-z = C p T + ty (3.2.3)The coefficients Kx and C x are assumed to be the same for u, v, q v , q w and ,v . Using m and h as sub-scripts for momentum and heat, the problem is reduced to the determination of K fn and K h (at all atmo-spheric layers) and C m and C h at the surface. This has to be done solely from atmospheric values (att - At) of u, v, T, q v , q w and from the surface conditions.Surface fluxesThe surface fluxes for momentum, dry static energy and cloud water are parameterized as follows:18J u = P C m' ' «J v = P • C m ■ '■?/, ■ V4 = P • • i * ■ <«-«s)J q . = P- c h - ii ■(3.2.4)where C m is the drag coefficient, C h the transfer coefficient for heat and v h the absolute value of thehorizontal velocity. C m , C h and |y fc - apply to the lowest model level at z - Ar. /<. v and 5 are the velocitycomponents and the dry static energy to be computed implicitly for the lowest model level. s s i s the drystatic energy at the surface at t - Ar, and q wS is the cloud water content at the surface (set to zero).The moisture flux is evaluated distinguishing sea and land. Over seaJ q, = P • C h • W • (3.2.5)where q s is the saturation specific humidity at surface temperature T s and pressure p s .Over land, each grid square is divided into 4 fractions:1. fraction CS n covered with snow2. fraction -C; covered with water in skin reservoir3. fraction ( l - c s „) • ( 1 -C j ) • ( 1 -C V ) covered with bare soil4. fraction (1 -C S „) • ( 1 -C , ) c v covered with vegetationThe snow cover fraction C Sn depends on the snow cover Sn :C Sn = wzn( l , — ) (3.2.6)ön crwhere Sn cr is the critical snow cover (= 0.015 m equivalent water depth).The wet skin fraction C z is derived from the skin reservoir water content:( W i }C, = m in i 1, —— (3.2.7)l w lmxJwithW lmx = HW(1- ) + C v -L,] (3.2.8)W l is the prognostic variable for the skin reservoir content, W lmx is the maximum skin reservoir con-tent, C v is the vegetation ratio, L t is the leaf area index, and W lmax is the maximum amount of waterthat can be held on one layer of leaf or bare ground. It is taken to be 2. • 10-4 m.The grid fraction C v occupied by vegetation is equal to the climatological field C vc l except in dry con-ditions when the vegetation is reduced according to the following empirical expression:19( 'IC v = mm C vcl , C vc l ■ - (3.2.9)k v.M- Smax JThe quantity represents the total amount of water available in the root zone and is the fieldcapacity (0.2 m).Evaporation from snow and the skin reservoir is at the potential rate= P 'C h -\yh - {q v -q s (T s , p s ) } (3.2.10)For the evaporation from bare soil (no water in skin reservoir) it is assumed that the relative humidity hat the surface is related to the water content of the soil layer:Jq yii) = P -C h \v - {q v -h q s (Ts , p s ) (3.2.11)K-W— \’ min (3.2.12)h = max \ 0.5 ■ 1 - cosThe evaporation from dry (no water in skin reservoir) vegetated areas is proportional to the evaporationefficiency E :Jq ,(.iv') = p ■ C h -\y ■ E ■ [ qv -q s (T s , p s ) } (3.2.13)Based on Sellers et al. (1986), E is expressed as:r c * m-wmi’* F(W S ) J (3.2.14)E =where R co /F (W s ) is the stomatai resistance of the canopy, with a minimum value R co dependent onthe Photosynthetically Active Radiation (PAR), and an empirical function of the available water in theroot zoneF(lV5 ) .1 _ 1 F b de kL 'Rco k -cLd-PAR ' " d+ 1d + ed + 1 (3.2.15)whered = f . k = 0.9, L ,= 4, a = 5000 Jm~ 3 , b = 10~Wm~ 2 , c = 100 sm' 1c ■ PAR 1and PAR is 55 % of the net short wave radiation at the surface. In case of dew deposition (q v > q s ) weset(3.2.16)E = h = 1The water stress factor F ( Ws ) is20if (Ws >w cr )1W S -* pwpvy sW - Wcr P w P0W pwp • ( s - \ v ) (3-2.20.C)Equations (3.2.19.a)and (3.2.19.b)canbe integrated over the lowest model layer, and L eliminated using(3.2.20.a) in order to derive C m and C h . However, such expression cannot be obtained analyticallybecause of the complicated form of and Cm and C h are approximated by following analy-21tical expressions (Louis, 1979):k 2C m = (7 - - -T7 “x ) ’ f m 2 2 (dIn ( z / z 0 ) 0(3.2.21. a)andk 2c ‘= W (3. 2.21. b)here the Richardson number Ri is defined as:g • Az • As (3.2.22)C ■T-lAu 2The empirical functions f m and f h must have the correct behaviour near neutrality and in the asymptoticcases of high stability or instability.a) Near neutrality one obtains Ri —> 0, z/L —> 0, 0 = 1 + b m • z/L and = 1 + b h • z /L .We obtain then(3.2.23)hFurthermore , there is some evidence (Pandolfo ,1967), that Ri ~ z/L,which implies b h = 2 • b m . Thus we have(3.2.24)In ECHAM b = 5.b) In highly unstable cases only the problem of f h in the case of vanishing u(free convection case) has been considered. To have a non-zero heat flux,u must vanish from (u - / A ) for large negative ( s - s s ) (that is for high negative value of Ri).So f h must behave like -Ri .The analytical expression chosen for the whole unstable case (R i<0) is:For this free convection case one can write an equation for a new similarity theory,independent ofz dss* dz(3.2.26)22with9 _ I/O(VP)5 * _ (g z ) / s_Using Eq. (3.2.21) for the limit Rz —> «>,J 37?c h (z /z 0 ) = C - 3 ■ b\ I [ (z/z 0 ) 1/3 - 1 | ' (3-2.27)in { z / ZqJ j1/3 -----------One can furthermore replace [ ( z / z 0 ) — 1] by ( z / z 0 ) if z is much greater than z 0 , so thatk - 2 . _______Ch ( z / z 0 ) = C-3 -b t 7 (z/z 0 ) (3-2.28)In order to avoid numerical problems for high z 0 values that might suppress z one replaces(z /z 0 ) by (z /z 0 ) + 1 in (3.2.21) and (3.2.27), (3.2.28).InECHAM C = 5.In the highly unstable case f m is not very important since there is little wind shear on which to act.Therefore a similar expression to (3.2.21) has been chosen with the same denominator to savecomputing time2 • b Ri1 ” r ’ , _21 + 3 ■ b • C \ . . * . . . \ J(z/z 0 +1 ) (-«i)\ In ( z / z 0 + 1) uc) Finally in the highly stable case we follow Ellison (1957) by combining two equations«>).J Li„ Z 1 'For y —» +oo this givesLtand4~ l / (7? i 3/2 )For lack of better information and to minimise the computing time for the wholestable range (Ri > 0) the following expressions have been chosen(3.2.29)(3.2.30)(3.2.31)(3.2.32)23m 1+ (2 - /> -R i ) / (v ' l+d -« i )f h = ------------------ ’------7==?=-- < 3 - 2 - 34 )1 + (3 b Ri) • (71 + d -R i )d is related to the critical flux Richardson number R cri t by2= f~ d (3.2.35)InECHAM d = 5 .Definition of the exchange coefficientsThe logical extension of the similarity theory for surface fluxes to the atmosphere is the mixing lengthapproach.K = l2 - \ l l h f m (Ri) (3.2.36)m m \~\ J y '|ozKh = l2 h \ - f h (Ri) < 3 ' 2 ' 37)oz IThe functional dependencies of f m and f h on Ri are the same as for C m and C h except that the term. _2! ___k- ___ !\ Jn ( z / z 0 )1/3 3/2( (z/z„) ZJ - 1]in (3.2.27) is replaced by its equivalent-3/2-1 i1/3__?(Az) 3/2z + Az(3.2.38)1/2zThe intensity of the vertical diffusion in the atmosphere is dependent on the choice of l h and lm . Here,the solution of Blackadar (1962) is used, that goes smoothly from the asymptotic value k ■ z for z —> 0to a constant value in the high atmosphere:1 = J_ J_ 1 = 1 1/ k • z + X ' l h k- z + X."i m n n(3.2.39)One also has to have a relationship between X , X h and d so that the critical flux Richardson numberin the atmosphere is 1 (e.g. a balance between shear generation and buoyancy destruction).Xm2R -,(z —> °°) = —7 3d (3.2.40)= 124gives \ = X. m -,'(3 • d ) /2 .InECHAM X = 160 m .mMoisture and cloud effectsIn the ECMWF model and also in ECHAM1 and ECHAM2, moisture effects on stability have beenincluded by replacing the temperature T in the definition in the Richardson number (3.2.22) by the vir-tual temperature T v since it is T v that determines the buoyancy in moist (but cloud-free) air. In ECHAMcloud effects are considered by reformulating the Richardson number to include the impact of cloud pro-cesses on the buoyancy term (Brinkop, 1992). This so-called moist Richardson number can be writtenaccording toQ A ■ A0.+ 0 ■ B • AqRim = w ----------- ----------2~ ' Az®V (Au) 2 + (Av) 2(3.2.41)where q t = qi + Qi + q v is the total water content, 0 v is the virtual potential temperature,© z = 0 - ( 0 . L • q /T • C d ) is the liquid water potential temperature which is a conservative quan-tity in the absence of freezing, precipitation and radiative effects (Betts, 1973), and 0 is the potentialtemperature.The parameters A and B are defined as follows:Unsaturated caseA unsal = 1+0 .61 • var ( NE / S W)z c is the atmospheric depth corresponding to three-quarters of a hydrostatic vertical gravitywavelength and is computed by the solution of the equation(z) _ 3ki l / ( z ) äz 20(3.3.3)The scheme can then be written as oj (p) = T w (p) + T Fr (p) .The first term (p) describes a part of the low-level drag and the upper part, dependent on thepressure ps Ps) • d -ß ) EEr for (p-p">PS Pß- J w (Ps) f(p) for (p
Fr c ) (3-3.5)PS Pz cwheref H 2X Fr (p s ) = P (Fr -Fr c ) f 2 (x) if (Fr>Frc )0 otherwisep z is the pressure corresponding to the height z c and KL is currently 4 K. Typical values of z care around 3 to 5 km but much larger values do occur.28(ii) The vertical structure of T gyv is calculated by constructing a local wave Richardson numberwhich attempts to describe the onset of turbulence due to the gravity waves becomingconvectively unstable or encountering critical layers.-1 /2 2This wave Richardson number can be written in the form R = / ? ( ] - a ) / ( 1 + a • /? )where R is the Richardson number of the basic flow. The parameter a = N ■ Öz/u i sa formof inverse Froude number in which | Sz represents the amplitude of the wave and u is the windspeed resolved in the direction of r gw .By requiring that R never falls below a critical value R cr i t (currently equal to 0.25 ), values of wavestress are defined progressively from the surface upwards.3.4 CUMULUS CONVECTIONCumulus convection is parameterized by a mass flux scheme which is described in detail (includingnumerical aspects) in Tiedtke (1989). The scheme considers penetrative convection, shallow convectionand mid-level convection. Clouds are represented by a bulk model and include updraft and downdraftmass fluxes. Momentum transport by convective circulations is also included following the proposal bySchneider and Lindzen (1976).Large-scale budget equationsThe contributions from cumulus convection to the large-scale budget equations of heat, moisture andmomentum are:+ L ( c u~ e d-~e i - ' e p') - (L s -L v )mCU = ~Ü + " A +cu = ~ lM “ U“ + M “ U “ ' + M“ ) “ ]( | ) cu = - [ M u v u + M d v d - (M u + M d ) ?](3.4.1)29where M u , M d , c u and e d are the net contributions from all clouds to the upward mass flux, downwardmass flux, condensation/sublimation and evaporation respectively, and s d , q u . q d , u u . u d , v u and v dare the weighted averages of 5, q v , u and v from all updrafts and downdrafts. Here e { is the evaporationof cloud water that has been detrained into the environment, ep is the evaporation of precipitation in theunsaturated subcloud layer and m is the melting of snow. In addition to (3.4. 1 ) we consider the equationsfor precipitationP" (z ) = f (G wp - e wd -~ep + m)pdzJ (3.4.2)P l (z) = J (G lp - e d - e P + m) pdzwhere P w (z) ,P l (z) are the fluxes of rain water and snow at height z and G"p and G lp are the conversionrates from cloud ice and cloud water into precipitation, respectively.Cloud model equationsThe updraft of the cloud ensemble is assumed to be in a steady state. Then the bulk equations for mass,heat, moisture, cloud water content and momentum are— = E -Ddz u uf z (M u s u ) =E u s -D u s u + Lpc u=E u q v -D u q u -pc udz (3.4.3)| (MJ) = D u l + pc u -pG P= E u“~ D u u uW= E uV- D u v uwhere E and D are the rates of mass entrainment and detrainment per unit length, / is the cloud watercontent (water/ice) and c u is the net condensation/sublimation in the updrafts.Cloud air is assumed to be saturated and cloud processes are crudely represented. Freezing and meltingprocesses are not considered and the conversion from cloud droplets to rain/snow is assumed to be pro-portional to the cloud water content aspG P = K(z ) - l -M uwhere K (z) is an empirical function that varies with height. This simple parameterization yields ratherreasonable vertical distributions of the generation of raindrops (Yanai et al., 1973). Here K is assumedto be zero near cloud base and constant at higher levels300, if ( zZ ß + Az)(3.4.4)where Az is 1500 m over sea and 3000 m over land. The choice of K = 0 at lower levels ensures thatshallow cumuli do not produce precipitation, noting that a sizeable portion of the liquid water content innonprecipitating cumuli is of precipitation- sized drops.We further note that the cloud water detrained into the environmental air is assumed to evaporate thereinstantaneously. Thene . = =-D I (3.4.5)' p uThe vertical integration of (3.4.3) requires the knowledge of cloud base mass flux and mass entrainmentand detrainment. Cloud-base mass flux is determined for the various types of convection from theparameterization assumptions discussed below. Entrainment of mass into convective plumes is assumedto occur through turbulent exchange of mass through the cloud edges and through organized inflowassociated with large-scale convergence, detrainment through turbulent exchange and as organizedoutflow at the top:£„ = S* 1 ’ + £ ‘ 2) , D=D +D (u 2) (3.4.6)Turbulent entrainment and detrainment are parameterized as£„(1 ) = zM u , D‘” = 8„A/„ (3.4.7)where the fractional entrainment/detrainment rates depend inversely on cloud radii (Simpson andWiggert, 1969; Simpson, 1971):2 0.28 «= uBy assuming typical cloud sizes for the various types of convection, average values of entrainment/detrainment rates are defined. In the presence of synoptic scale flow convergence, large clouds whichcontribute most to the convective heating and moistening are assumed to exist and consequently smallvalues for entrainment/detrainment rates are imposed whereas in the absence of flow convergence cloudsof smaller sizes with larger entrainment rates prevail. In order to keep the scheme simple we use fixedvalues of turbulent entrainment/detrainment rates for each of the various types of convection:0.2£ “ Ru(3.4.8)for penetrative and midlevelconvection in the presenceof large-scale flow convergencefor shallow convection insuppressed conditions1 x 10 Si 1e = 8 = {u u L3 x 10 -4 w 1(3.4.9)For penetrative convection and midlevel convection we deliberately impose a very small value typical31for tropical deep convective clouds (Simpson, 1971) so as not to inhibit the penetration of clouds to largeheights. For shallow convection we use a value typical for the larger trade wind cumuli (Nitta, 1975),noting that small clouds with much larger entrainment/detrainment rates which detrain immediatelyabove cloud base are not represented in our parameterization.The parameterization of organized entrainment and detrainment is discussed below.Below cloud base, the net convective fluxes of heat and moisture due to updraft and compensatingsubsidence in the environment are assumed to decrease linearly from their values at cloud base towardszero at the surface so as not to alter the vertical structure of the subcloud layer and in particular its well-mixed character.Downdrafts are considered to be associated with convective precipitation from the updrafts and originatefrom cloud air influenced by the injection of environmental air. Following Fritsch and Chappell (1980)and Foster (1958), the Level of Free Sinking (LFS) is assumed to be the highest model level where amixture of equal parts of cloud air and saturated environmental air at wet-bulb temperature becomesnegatively buoyant with respect to the environmental air. The downward mass flux is assumed to bedirectly proportional to the upward mass flux. Following Johnson (1976, 1980) the mass flux at the LFSis specified from the updraft mass flux at cloud base as(M d ) = with y = -0.3 (3.4.10)Lj J O (zCX jC-The coefficient y is a disposable parameter.IThe vertical distribution of the downdraft mass flux, dry static energy and moisture below the LFS isdetermined by the equations for mass, dry static energy and moisture content as=E ‘ - D “= E d ~ D d S d + E P e d= E d v~ D d‘ld + P e d (3.4.11)-£- z (M d u d ) = E d u-D d u d~ M d v d> = E d i ~ D d v dWe note that cumulus downdrafts can be viewed as the reverse of updrafts. Entrainment and detrainmentin downdrafts are highly uncertain as relevant data are not available. Numerical experiments show,however, that the results are rather insensitive to changes in the entrainment and detrainment rates. Weusee d = 8 d = 2 x l()’ 4 m“' (3.4.12)32This gives a mass flux which is independent of height and which effectively detrains into the environ-ment in the subcloud layer. We also note that e d is the evaporation of convective rain to maintain a satu-rated descent and that the moistening and cooling of the environmental air injected at LFS is also due toevaporating rain. As the downdrafts are determined from the updrafts the remaining parameterizationtask is to specify the updraft. This is done by means of closures defined below for the various types ofconvection.a. Penetrative convectionMany diagnostic studies show that penetrative convection predominantly occurs in disturbed situationsand strongly depends on low-level synoptic scale convergence. Various parameterization schemes arebased on this relationship one way or another. Here, we apply a moisture convergence hypothesis:Following Kuo (1965, 1974) and Lindzen (1981), we postulate that when there is a deep layer ofconditional instability and large-scale moisture convergence, cumulus clouds exist that entrainenvironmental air through their base and through their sides directly proportional to the supply ofmoisture and detrain cloud air at higher levels.The injection of mass into the clouds through their base is determined by imposing a moisture balancefor the subcloud layer such that the moisture content is maintained in the presence of large-scaletransports, turbulent transports and convective transports. This balance may be written asfif F_ _ _dq v 1 9 (P h ' , ?'v) | u\M u (q u ~q v ) +M d (q i -q v )] B = J l y h - Vq v + w += pdz (3.4.13)owhere B denotes the cloud base height defined as the condensation level for surface air. The verticaldistribution of the updraft mass flux above cloud base is determined using similar arguments as for thesubcloud layer, that is, we postulate that there is organized entrainment which is directly proportional tothe large-scale moisture convergence as£ « 2> =- L ' v + (3414)Organized entrainment is only considered in the lower part of the cloud layer where large-scaleconvergence is encountered, that is, below the level of strongest vertical ascent. The idea to link the cloudmass flux directly to the large-scale moisture convergence Jias first been advocated by Lindzen (1981)who indicated that it may provide vertical profiles of mass flux and convective heating in good agreementwith observations. The assumption (3.4.14) ensures that the vertical distribution of the convective massflux follows that of the large-scale ascent which is partly supported by diagnostic studies for tropical con-vection (e.g. Cheng et al., 1980; Johnson, 1980). Equation (3.4.14) forms, together with the assumption(3.4.13) for the cloud base mass flux, the basic closure and as such is crucial for the performance of theparameterization of penetrative convection. The verification of the scheme during long periods of tropi-33cal convection confirms that the closure provides realistic profiles of convective mass fluxes and convec-tive heating (see Tiedtke, 1989).In addition to organized entrainment we consider turbulent entrainment and detrainment by equations(3.4.7) and (3.4.8).Cumulus clouds detrain effectively at levels near to their zero-buoyancy level by means of organizedoutflow. Therefore, the vertical distribution of the total detrainment from all clouds depends on thespectral cloud distribution. Since spectral cloud distribution is not available, however, organized outflowis assumed to occur only in the model layer which contains the zero-buoyancy level of the deepestclouds. Our detrainment assumption implies an unimodal cloud distribution with large detrainment fromthe deepest clouds and little detrainment from shallow clouds and medium deep clouds.The effect from shallow cumuli in the presence of penetrative convection has been neglected becausetheir parameterization is still an unsolved problem. This is because the role of shallow cumuli inconnection with penetrative convection is not well understood, particularly when cumulus downdraftoccur simultaneously as these compete with shallow convection having similar effects on theenvironment as shallow cumuli (Johnson, 1976). The results obtained with this scheme indicate,however, that neglecting the contributions from shallow cumuli when penetrative convection isencountered does not introduce large errors in the convective heating and drying.b. Shallow convectionHere we consider cumulus convection, which predominantly occurs in undisturbed flow, that is in theabsence of large-scale convergent flow. Typical examples are trade wind cumuli under a subsidenceinversion, convection occurring in the ridge region of tropical easterly waves and daytime convectionover land. This type of convection seems to be effectively controlled by subcloud layer turbulence. Infact, most of the diagnostic studies carried out for tradewind cumuli show that the net upward moistureflux at cloud base level is nearly equal to the turbulent moisture flux at the surface (Le Mone and Pennell,1976). As this implies a quasi-steady moisture balance, we shall apply the same moisture budget equa-tion (3.4.13) as for penetrative convection. The difference, however, is that the moisture supply tocumulus clouds is now largely through surface evaporation as the contributions from large-scaleconvergence are either small or even negative, such as in the undisturbed trades where dry air istransported downward to lower levels.Under typical tradewind conditions the vertical distribution of the total -convective fluxes above cloudbase is dominated by two types of clouds: very small cumuli, which in large -numbers detrainimmediately above cloud base, and deeper clouds, which detrain just beneath and above the tradeinversion. The intrusion of cumulus clouds into the stable layer above the inversion is throughovershooting of cumuli above their level of zero-buoyancy (Nitta, 1975). Because of the coarseresolution employed in large-scale models, where the vertical gridlength is typically 50 to 100 hPa, it isdifficult to represent these two types. The scheme presented here ignores the effects of very small cumuli34but tentatively accounts for the effects of overshooting cumuli, as we assume that a given fraction of thecloud ensemble penetrates into the inversion layer and detrains there into the environment. Thus, cloudair shall only partly detrain into the environment within the model layer that contains the zero-buoyancylevel; the remaining fraction shall intrude into the next layer above and detrain there:~(2) _ jt+1/2u ~ \zn (2) ß “ \+ l /2M Azfor top layer k(3.4.15)for layer k - 1where ß is a tunable parameter (currently 0.15)Although this parameterization is very crude, it clearly reproduces more realistic trade inversions thanwhen the effect from overshooting cumuli is ignored (ß = 0 ) , as then the inversion becomes too strongin the simulation and the cloud layer below the inversion too moist (see Tiedtke, 1989).c . Midlevel convectionMidlevel convection, that is, convective cells which have their origin not in the boundary layer but startat levels above the boundary layer, often occur at rainbands at warm fronts and in the warm sector oftropical cyclones (Browning et al., 1973; Houze et aL, 1976; Herzegh and Hobbs, 1980). These cells areprobably formed by the lifting of low level air until it becomes saturated (Wexler and Atlas, 1959) andthe primary moisture source for the clouds is from low-level large-scale convergence (Houze et al.,1976). Often a low-level temperature inversion exists that inhibits convection from starting freely fromthe surface; therefore convection seems to be initiated by lifting low-level air dynamically to the levelof free convection. This occurs often in connection with mesoscale circulations which might be relatedto conditionally symmetric instability (Bennetts and Hoskins, 1979; Bennets and Sharp, 1982) or awave-CISK mechanism (Emanuel, 1982).Although it is not clear how significant the organization of convection in mesoscale rainbands is for thelarge-scale flow a parameterization should ideally account for both convective and mesoscalecirculations. Such a parameterization, however, is presently not available and we must therefore rely onsimplified schemes. Here we use a parameterization which in a simple way considers the finding of thediagnostic studies mentioned above. We assume that convection is activated when there is a large-scaleascent at lower levels, the environmental air is sufficiently moist, i.e., of.relative humidity in excess of90 %, and convectively unstable layer exists above. The free convection level is determined by lifting aparcel of environmental temperature and moisture contentT U = T , q u = q v (3.4.16)adiabatically, allowing for condensational heating, and then checking for buoyancy. The upward massflux is set equal to the vertical mass transport by the large-scale flow at that level:35(M u ) b = P B W B (3.4.17)which ensures that the amount of moisture which is vertically adverted through cloud base by the large-scale ascent is fully available for the generation of convective cells.In addition to the injection of mass through cloud base, we assume again that cloud air is produced bymo: ure convergence above cloud base through lateral entrainment in the same way as for penetrativeconvection given by equation (3.4.14).Evaporation of rain and melting of snowMelting of snow is parameterized as stratiform precipitation (see section 3.5 )The evaporation of convective rain is parameterized following a proposal of Kessler (1969), where theevaporation is assumed to be proportional to the saturation deficit q s - q v and to be dependent on thedensity of rain M R (g /m )E = a r ( 9 j - ? v ) -< 3 /2 ° (3.4.18)where a } is a constant being zero for q v > q s .As the density of rain M R is not given by the model it is convenient to express it in terms of the rainintensity R [g / (m sec)] as(3.4.19)7? — M R ( V q 4- w ) = M r ■ Vqwhere Vo is the mean fall speed of rain drops which again is parameterized following Kessler (1969).. , 1 / 8a 2 ' Mr■Jp Ps(3.4.20)0 =Thus we have-------- 13/20-i 8 /9E = (3.4.21)«1 ■ (<7s - ?y )Since the convective rain takes place only over a fraction C c of the grid area the evaporation rate at levelk becomesi------------- _ 3P r „ , i f jP'Ps R 'IE = C - a ■ ( q — a ) \ -------- • —c “ 1 C Jwhere the constants have the following values (Kessler, 1969)a , = 5.44 • IO’4 , a 2 = 5.09 ■ 10’ 3 , a 3 = 0.5777In order to save computing time (with a 3 = 1 /2 ) we use slightly differe- alues04 = 6.94 • 10" 4 , a 2 = 7.35 • 10 -3 , 04 = 0.5(3.4.22)36In view of the uncertainty of the fractional area of precipitating clouds a constant value ofC c = 0.05is assumed.The evaporation rate is calculated implicitly in the model by means of2g -R 1 /2 = -A (3.4.23)OPwhich follows fromr a «1/2 a z x PS 1 z , AE = AR A = a } (q s -q v ) ■ <3.4.24)andStratocumulusStratocumulus convection is parameterized by means of a vertical diffusion scheme (Tiedtke et al.,1988). It is only applied in the boundary layer where it provides a net upward transport of moisture andcloud water to avoid the generation of saturated layers wi±in a convective boundary layer.The net effect of stratocumulus convection is given by the turbulent fluxes of heat, moisture and cloudwater, which are parameterized on the basis of the eddy diffusion theory (cf., section 3.2):Sv H ( > K & 0 ) , the evapo-ration of cloud water C < 0 , the formation of precipitation by coalescence of cloud droplets and sedi-mentation of ice crystals (P > 0) , and the evaporation of precipitation in unsaturated air ( £ > 0 ) .Sub-grid scale cloud formationSince real clouds are often smaller than the size of a grid box, sub-grid scale cloud formation is takeninto account. The formalism has been developed by Sundquist (1978). Assuming that a fractionalhorizontal area (b ) of a grid box is covered by clouds while the remaining part ( 1 - b) is cloud-free,the equations (3.5.1) and (3.5.2) are modified according toy/ = -b C c - (1 -7? ) -C o + (1 - / ? ) ■ E o (3.5.3)öo= R(q„)+bC c +( \ -b ) -C ' -b -P , (3.5.4)where the subscript ( c ) denotes the respective process in the cloudy part and the subscript (o) theprocess in the cloud-free part of the grid box. The transport terms are assumed to be identical in bothparts as well as temperature and wind. Evaporation of precipitation is not allowed in the cloudy part, i.e.E c = 0 , and there is no precipitation formation outside the cloud, i.e. P o = 0 . The significance of C o ,38which is formally retained in eqs. (3.5.3) and (3.5.4) will be discussed below. In the cloudy part, satura-tion with respect to the grid averaged temperature T is assumed, q c = q s ( T) , while the cloud-free partis unsaturated by definition, i.e. q o < q s (T) . Hence, fractional cloud cover (/?) and grid-mean watervapour mixing ratio (q v ) are related according to( 1 - * ) - Qo (3.5.5)or, in terms of the relative humidity ( r = q v / q s ) and solving for (b )r - rb = t — - (3.5.6)X ~ r ofor r> ro and b = 0 otherwise. The relative humidity in the cloud-free part ( /;,) has to be specified orparameterized. The choice of r o should depend on factors such as grid resolution and sub-grid scalevariance of vertical velocity (Sasamori, 1975). Presently, we specify r o as a function of height andstability only: In stable stratification, r o decreases linearly from the surface layer ( r ? = 0.99) to the topof the PBL (r o = 0.85) above which r o remains constant. If penetrative convection occurs, r o decreasesfurther to a minimum value of 0.5 near the tropopause , as suggested by Xu and Krüger (1991) whoshowed in a modelling study of a cumulus ensemble that convectively driven stratiform clouds such ascirrus anvils may form already in a relatively dry environment.Condensation and evaporationCondensational growth of cloud droplets is assumed if the grid-mean relative humidity r exceeds thethreshold r0 . Oppositely, an existing cloud will be diluted by evaporation if r < r o . In the case of cloudformation by moisture convergence or adiabatic cooling, for example, any supersaturation in the cloudypart of the grid box will result in a condensational growth of cloud droplets.A basic problem is to specify the partitioning of the net moisture convergence between the /)-part andthe ( 1 - b) -part of the grid box.dqvAccording to the definition (3.5.5), 5- may be decomposed into three parts,d rdr b 'dt (1 b) 'dr q < q 'dt (3 ' 5 ’ 7)which represent the changes of q s , q o and Z due to temperature changes and moisture convergence.Inserting (3.5.7) into (3.5.3) and separating the b -terms, provides the condensation (evaporation) rate C cas a function of moisture convergence (divergence) and adiabatic cooling (heating).b- = b -R(q v ) - bC c (3.5.8)According to (3.5.8), the b -fraction of the moisture convergence will be used for condensation whilethe remaining part, ( 1 -b ) -R (q v ), increases the relative humidity in the cloud-free part. The39condensation rate is calculated from the moisture convergence into pre-existing clouds only.Furthermore, we assume that there is always an abundance of cloud condensation nuclei and ice nucleiso that condensational growth is allowed to start as soon as the 100 % relative humidity threshold in theb-part of the grid box is exceeded. The saturation water vapour pressure is calculated from Tetens for-mula (Lowe, 1977) with suitable coefficients for the liquid phase and ice phase, respectively.It remains to calculate the cloud water evaporation rate (C o ) in the cloud-free part of the grid box. Cloudwater outside the Z?-part may be generated by advective or diffusive transports across the boundaries ofthe grid box while internal mixing by molecular or turbulent diffusion is neglected. Moreover, numericaleffects do contribute also as, for example, the representation of cloud water by spherical harmonics inthe host model. In these cases we assume that cloud water in the (1 - b ) -part will be evaporatedinstantaneously,R(q w ) +C o = 0 (3.5.9)Note that all numerical effects are included in the transport term R (q w ) which may also becomenegative. In this case, C o is positive and actually represents a condensation rate.Precipitation formationAnalogous to equation (3.5.5), the grid-mean cloud water mixing ratio is given byQ w = bq c (3.5.10)assuming that all of the cloud water is confined to the cloudy part of the grid box which implies that thein-cloud mixing ratio is defined as q c = q w / b .The mechanism of precipitation formation depends crucially on the cloud water phase. Since we applyonly one budget equation (3.5.4) for the cloud water mixing ratio, the liquid and ice phases are diagnosedas a function of temperature T according to+ = Qe- f iW +q c fi(T) (3.5.11)with +/■ = 1 . The fractions and f- have been obtained from a wealth of aircraft measurements, ascompiled by Matveev (1984), by applying an exponential fit to these data (Rockel et al., 1991),/ ( (T) = 0.0059 + 0.9941 e' 10003102 (T ' 273 15 r l (3.5.12)The growth of cloud droplets to precipitating rain drops by autoconversion is .parameterized in aconvenient exponential form (Sundquist, 1978). In addition, the collision of cloud droplets with largerrain drops is taken into account (Smith, 1990) so that the total coalescence rate is given byPci = ‘ l c l lCr ib - ' / q \ +C r2 (P)j (3.5.13)where (P) is the rain flux density (stratiform and convective) at the top of the respective cloud layer and40C r l , C r2 and q cr are microphysical constants which determine the efficiency of rain formation and,thus, the cloud lifetime (in ECHAM3/T42 C r l = 10 -4 s -1 , C r2 = 1 m 2 kg~ ] and q cr = 3- 10 -4 ).Ice crystals settle at a rate which depends on the form and size of the crystals. Both parameters are notavailable in the model. However, according to an observational study by Heymstield ( 1 977), the terminalvelocity of the crystals can be parameterized in terms of the ice water mixing ratio which is a modelvariable,V, = a . ( p a <7c; ) ß (3.5.14)where p a is the air density, and the constants a and ß are obtained from a best fit to the data with a slightadjustment (tuning) in the respective model version (in ECHAM3/T42, a = 1 777 and ß = 0.16). Theloss of ice crystals due to sedimentation is given by the divergence of the ice flux density,Pci = g (3.5.15)where p is the pressure and g is the acceleration of gravity. The total rate of precipitation formation isgiven byP c = Pd + Pci (3.5.16)A crucial assumption in the parameterization (3.5.13) - (3.5.16) is that the precipitation formation in amixed phase, i.e. in a temperature range between about 0 0 C and -40 0 C, can be treated independentlyfor the ice phase and the liquid phase, respectively. A proper treatment of the interaction between bothphases, such as the rapid condensational growth of ice crystals at the expense of cloud droplets(Bergeron-Findeisen process), will require a more elaborate scheme which should be based on thebudget equations for each phase.Evaporation of precipitation and melting of snowPrecipitation falling into the cloud-free part of a grid box is exposed to evaporation which isparameterized in terms of the saturation deficit according to1 Y-Ar L 'dq s /dT1 . + c pd - (1 + (Ö-"1)<7 V )(3.5.17)where y is a tunable parameter (currently y = 0.1 ) and 8 = C pv /C pd . The evaporation rate in a layerwith thickness Ap is limited, however, by the precipitation flux density at the top of the layer, where ( )denotes a vertical integral of the respective quantity,(E o (p + Ap) )< (P (p)> (3.5.18)41with = -J [* Pc ~ (1 ~b)E o ]dp‘ (3.5.19)oand(P + Ap)(£ o (p + Ap) ) = J Eo dp' (3.5.20)opMelting of snow is parameterized considering observational data summarized by Mason (1971). Meltingoccurs in a thin layer of a few 100 m below the freezing level. We therefore assume that the snow canmelt in each layer whenever the temperature exceeds 2 0 C. The melting is limited not only by the snowamount but also by keeping the included cooling of the layer such that the temperature of the layer aftermelting is not less than the 2 0 C threshold.3.6 SOIL PROCESSESThe surface parameterization scheme comprises the evolution of a temperature profile in the soil, the soilhydrology and the snow pack over land.Soil temperatureThe soil is divided into five layers according to Table 5. The thicknesses of the individual soil layersincrease with depth.Table 5 Thickness of individual soil layersLayer 1 2 3 4 5Thickness [m] 0.065 0.254 0.913 2.902 5.700This is an extension of the scheme of Warrilow et al. (1986) to five layers. At the lowest layer boundarya zero heat flux condition is prescribed in order to ensure that no artificial heat sources and sinks mayaffect the energy balance of the earth-atmosphere system. The heat conduction equation follows from/layer 1)(3.6.1)(layers 2 to 5)dTj _ F s 2k(T2 -T 1 )dt p g C g Az } ÄZj (Az j 4- Az 2 )2K(T- - r i _ 1 ) 2k( t - +1 - t 1.)d t Az : (Az-_ j +Az t ) Az- (Az- + Az- + j)with42K heat diffusivity in the soil 7.5- 10‘ 7 irrs' 1p • C heat capacity of the soil per unit volume or 2.4- 10 6 J m ' 3 K 1& a 7 — 1heat capacity of land ice per unit volume 2.09- 10 J m KF s sum of the radiative and turbulent fluxes at the surface if there is no snow;heat flux from the snow to the deep soil if the snow depth exceeds 0.025 m.Snow pack temperatureThe scheme accounts for the three different conditions at the surface boundary in the presence of a snowpack over land:1) In the case of permanent ice cover at a grid cell, the soil heat equation Eq. (3. 6. 1 ) is solvedassuming the characteristics of ice. These areas are defined by the glacier mask.2) For a snow pack of less than 0.025 m water equivalent (3.6.1) is solved assuming thecharacteristics of the bare soil.3) For snow depths deeper than 0.025 m, an extra heat conduction equation (Bauer et al., 1986)evolves according to3 7- = F sdl S " Psn ■ C sn • Sn(3.6.2)withT Sn temperature in the middle of the snow packFs sum of the radiative and turbulent fluxes at the surfacep s „6?Sn heat capacity of the snow per unit volumecomputed using a density of snow p 5n ofSn depth of the snow pack0.6345- 106 J m’3 K“ 1300 kg m’3Snow meltThe temperature T Sn is used to define a temperature T s at the top of the snow pack by extrapolation. T sserves as an interface to the atmosphere. This temperature may not exceed the snow melt temperatureT melr If sufficient energy is available to raise the temperature T Sn above T melr this energy is used towarm the soil underneath. Only if both T\ and T Sn reach the value of T mell , further energy input will beused to melt the snow. Snow that is less deep than 0.025 m may melt if equals T melrSea-ice temperatureThe ice-surface temperature T ice is calculated from the energy balance at the ice surface. To avoidproblems due to a large time step it is assumed that T ice represents the uppermost 10 cm of the ice sheet.The energy balance equation for this layer can be written as43C p i ( Zt iCe ) F a , m + ce <3.6.3)where C p i is the heat capacity of a 10 cm ice layer. F a[m is the atmospheric heat flux consisting of thesolar and thermal radiation and the sensible and latent heat flux. The oceanic heat flux F oce isparameterized asF oce = A’ ( T oce~ T lC e ) (3.6.4)icewhere a is a heat transfer coefficient (2 W/mK), h ice is the ice thickness and T oce is the temperature ofthe underlying ocean.To solve equation (3.6.3) the atmospheric heat flux is linearised with respect to the surface temperatureat the previous time step T°s ldP a!m (T ice ) = F a tm (7? d ) + F atm (T i ce - T°s u ) (3.6.5)This leads to a linear equation for the new ice-surface temperature T ewrpold j. pi rpold’m 1 S JT 1 oce ' a f 1 ice, l ice<3P , « L a tmM h l c e dTatmnewS ~ (3.6.6)Soil hydrologyThe parameterisation of soil hydrology comprises three budget equations forsnow amount Sn (m water equivalent) accumulated at the surface.water amount W t intercepted by the vegetation during rain or snow melt episodes(the so-called skin reservoir),soil water amount VV'iii.The water equivalent of the snow layer is computed over land and glacier areas froma s„-M s „P»-Sn = —d t(3.6.7)withevaporation rate per unit area over the snow packsnow fall rate per unit areasnow melt rate per unit areadensity of waterRain water and melting snow on the leaves is intercepted by the vegetation until its water holding?SnM SnPw44capacity W lmx (cf. equation (3.2.8)) is exceeded. The corresponding budget equation is given bydW { J q vi + i p ' + 5«)V = -------- --------------------- ----- (3.6.8)dt p„withJ evaporation rate from the skin reservoirTv iP R rainfall rate per unit areaC v fraction of the grid box covered with vegetation (cf., (3.2.9))C ip coefficient of efficiency of rain and snow melt interception (currently 50 %)C a firactional area wetted by rain during a time step(currently 100 % for large-scale rain and 50 % for convective rainThe amount of rain and snow melt which does not enter the skin reservoir is used to calculate the amountof soil infiltration and surface runoff. The soil water reservoir evolves according toJq v , + P R P Ri + M Sn M Sn i~ Rd t ~ p> w(3.6.9)withJ grid-mean evaporation rate per unit area according to (3.2. 1 8)P Ri rainfall rate per unit area intercepted by the skin reservoir■ snow melt rate per unit area intercepted by the skin reservoirR r surface runoff rate per unit area from precipitaton events and snow meltR d runoff rate per unit area from drainage processesThe computation of R D and R r follows the scheme by Dümenil and Todini ( 1992) which is based oncatchment considerations. The scheme accounts for the heterogeneity of a grid area by assuming that thetotal field capacity W Smax for the grid area results from the integral of the local field capacities whichare distributed over the grid area in a non-linear way. The resulting fractional saturated area s /S is afunction of the degree of the grid-mean relative soil wetness W s /W Smax an< a s t ruc tu re parameter b thatdefines the sub-grid scale characteristics of the basin or grid box:5 ( Y- = 1 - 1 --------—S I J\ Smax /(3.6.10)Equation (3.6.10) defines a fractional saturated area s/S of a grid box where runoff would occur for acertain rainfall (or snow melt) event, while in the area 1 - ( s /S ) the rainfall would infiltrate (cf., Figure5). The surface runoff rate R r is computed from the area integral of rain and snow melt that arrives inthe saturated part s/S of the grid area. The amount of surface runoff resulting from a rainfall (or snowmelt) event during a time interval Ar (e.g. the model time step) is computed from45I 1 + b1 '"7' ( w $ }' +b QJ R R dt - Q (W Smax W s )+W Smax - 1 w j ( i +/,) . w'» w \ 3 max J ?> maxif [...] > 0 or± J R R dt = Q- (W Smax -W s )P w ;‘0if [...] <0 and Q + W s >W Smaxwheren '" 7 R- P Ri + M Sn~M Sni > t2 = ------------------------------ a tJ P„'ois the total water available for infiltration and runoff after possible interception in the skin reservoir (cf.,equation (3.6.8)).Equation (3.6.12) represents the well-known “bucket model” (Manabe, 1969) where runoff is computedif the precipitation event would cause an oversaturation of the whole grid box (i.e., + Q >Note also, that the scheme (3.6.11) converges to the bucket model for b —> 0 .In the ECHAM model, the structure parameter b is parameterized in terms of the sub-grid scale heightdistribution which is taken as a measure of the typical steepness of the terrain in the respective grid box:(3.6.11)(3.6.12)(3.6.13)"0 , - 0 “b = max\ ------------ ;0.01 i (3.6.14)1 & i, "h Ih max Jwhere o h is the standard deviation of the terrain height and O ö and a are prescribed minimum andmaximum values of o depending on the model resolution (currently, o rj = 100 m andmax ~ 1000(1500) m forT21 (T42) resolution).According to the runoff parameterization (3.6.1 1) - (3.6.14), surface runoff is extremely efficient in steepterrain (b approaching 0.5 for o A —> max ) while most of the precipitation is allowed to infiltrate the soilif the terrain is relatively flat (b —> 0.01 ).The infiltration rate per unit area IR is defined asj R R dtl 0(3.6.15)For frozen soil, however, we assume zero infiltration so that the surface runoff results as46— j R R dt = Q (3.6.16)Pw 'oRunoff due to drainage processes occurs independently of the water input Q if the soil wetness is bet-ween 5 % and 90 % of the field capacity (slow drainage) or larger than 90 % (fast drainage):(Y Smaxws ( ws -w ir \ dhU ■ wA + Umax - d mia ) U - W if tW'.V 2 w d,■>Smax \ V Smax ™ dr J(3.6A1)Pwwithdmin = 2-8 ' IO’ 10 m/s , d max = 2.8 • 10’ 8 m/s , d = 1.5, = 0.05 ■W dr = 0.9 • W SmaxThe amount of evaporation is computed from the atmospheric demand but is limited by the soil moistureavailability over bare soil and due to vegetation.heterogeneousdistribution of waterfield capacitiesnf thp enilW.■O TJevent 1 event 2fractional area s/SRUNOFF 1 INFILTRATION 15 ] INFILTRATION 2RUNOFF 2Figure 5 Surface runoff and infiltration from precipitation events.474. MODELS AND EXPERIMENTSA survey of the various model versions used in the present intercomparison study is presented in Table6. It also provides information on the design of the model experiments. In each climate simulation a cli-matological annual cycle of sea surface temperature (SST) is prescribed as boundary forcing whileinterannual SST variability has been neglected. It is important to note, however, that the SST data setsused in the various model simulations differ with respect to both, analysis technique and time period.All models have been initialized with the same set of atmospheric data (ECMWF analysis for January 1 ,1983), and they need a spin-up time of at least one year in order to settle to their own climate, respectiv-ely.Most of the observational data used for model validation are based on ECMWF analyses for a ten-yearperiod (i.e., June 1981 to February 1991). However, all zonal means and the eddy statistics (section 5.2)have been computed for a shorter period (i.e., June 1982 to February 1989).Table 6: ECHAM model versions and model experiments(* denotes the same formulation as in the previous version)ECHAMO(=ECMWF, cycle 17)ECHAM1 ECHAM2 I ECHAM3Dynam ics/numerics1Spectral (Baede et al.,1979)*, but revised formula-tion of the pressure gra-dient term (Simmonsand Chen, 1991)* *iJT42L19and (T21L19)Resolution T21L16 T21L19 *Timestep(min)45 40 * 24 (40)Horizontal diffusion1Linear, 4th order enhanced scale selec-tivity (Laursen and Eli-asen, 1989)* *- !Radiation Geleyn andHollingsworth (1979)Hense et al. (1982) andRockel et al. (1991)* *diurnal cycler ■no yes * *annual cycle yes * * *Vertical diffusion Louis (1979) *, but low wind correc-tion by imposing a min-imum wind speed forthe exchange of heatand moisture in unsta-ble conditions_________________* *, but low wind correc-tion according to Milleret al. (1992). The Rinumber is revised toinclude "moist" (cloud)effects48Gravity wave drag no Palmer et al. (1986) no Miller et al. (1989); noGWDatT21 resolu-tion.Convection Kuo (1974), no shallowconvection*, but with shallowconvection (Tiedtke etal., 1988)*•Tiedtke (1989) fordeep, midlevel andshallow convection.Stratocumulus convec-tion according toTiedike et al. (1988).Stratiform cloud Diagnostic scheme(Geleyn, 1981)Prognostic scheme(Roeckner andSchlese, 1985)* Prognostic scheme(Roeckner cl al., 1991)Soil processes Three-layer model forheat and soil moisturewith climatologicallyspecified deep soil tem-perature and moisture;no vegetation effects;sea ice temperature isclimatologically pre-scribed.Five-layer model forheat transfer; refinedbucket model for soilmoisture (Diimenil andTodini, 1992); vegeta-tion effects included(Blondin, 1989).* *, but sea ice tempera-ture is calculated fromthe net heat lluxesincluding conductiveheal transfer throughice.Orography Average over grid area("mean").Enhanced ("envelope") mean *Sea surface tempera-ture and sea ice extentAlexander and Mobley(1976)* for sea ice, but Rey-nolds (1988) for SST.* AMIP-SST and sea-icedatasetSimulated years in"control mode"(annual cycle of SSTand sea ice climatolog-ical ly specified; nointerannual varia-tions).10 23 23 30(10forT21)Averaging period(simulated years)2-10. . . .4-23 4-23 11-20; 21-30(2-10 for T21)Write-up interval(hours)24 12 12 12Table 6 (cont.)495. MODEL CLIMATOLOGYIn this section we compare the time-mean climate state, its intraseasonal variations and stationary eddyvariances and fluxes as simulated by different versions of the ECHAM model. Most of the discussionswill concentrate on the zonal averages of basic climate variables such as temperature, wind, pressure andprecipitation. We refer, however, also to the respective geographical distributions which are summarizedin Appendix A. The results of the T21 model simulations (ECHAM1, ECHAM2 and ECHAM3) are pre-sented for the time period indicated in Table 6 (section 4), whereas the results of the ECHAM3/T42simulation are shown separately for the first 10 years and for the second 10 years of the total simulationperiod of 20 years in order to demonstrate the stability of the statistics presented in this report.5.1 TIME-MEAN CLIMATETemperatureOne of the most apparent deficiencies of the ECHAMO model is a cold tropospheric bias of approxima-tely 3K globally with peak values of 7- 8K in the tropical upper troposphere (Diimenil and Schlese,1987). This error is caused primarily by a lack of latent heat exchange between the ocean surface and theatmosphere, predominantly at low latitudes. Another contributing factor is the excessive radiative coo-ling caused mainly by tropical cirrus cloud being optically too thin (Roeckner et al., 1989).In the more recent model versions (ECHAM1 to ECHAM3), the temperature errors are significantlyreduced, as apparent from Figs. 6 and 7, where the zonal mean errors (with respect to ECMWF analyses)are shown for northern winter (DJF) and northern summer (JJA), respectively. The most striking modelerror is the cold bias in the polar upper troposphere and lower stratosphere which prevails nearly inde-pendent of season, hemisphere or model physics and horizontal resolution. This common GCMdeficiency (e.g. Boer et al., 1991) is possibly caused by an insufficient vertical resolution of the stra-tosphere. The warm bias in the upper tropical troposphere which is typical of all models shown in Figs.6 and 7, but less common in other GCMs (Boer et al., 1991), is very likely related to deficiencies of theradiative transfer scheme (cf. section 3.1). As compared to "exact" line-by-line radiative transfer models,the present scheme systematically underestimates the (clear-sky) longwave cooling in the upper tro-posphere, while the longwave cooling in the lower layers is slightly too strong, particularly in the tropics(not shown). This feature is consistent with the vertical structure of the temperature error as shown inFigs. 6 and 7.All model versions perform reasonably well in the middle and lower troposphere, but the ECHAM3errors (at both resolutions) are slightly smaller in the low-latitude boundary layer. This is apparent par-ticularly in the Figs. A l and A2 (Appendix A) which show the respective geographical distributions ofthe temperature errors at 850 hPa for both seasons. Additional test simulations (not shown) with indivi-dual processes switched on and off have indicated that the error reduction is caused primarily (70%) by50the new mass-flux convection scheme in ECHAM3 (cf. section 3.4), while the remaining 30% is due tothe introduction of a "moist" Richardson number (cf. section 3.2) which allows for more efficient turbu-lent mixing within a cloud-topped planetary boundary layer.At 850 hPa (Figs. A l , A2), all model versions have a cold bias over the polar regions in both hemispheresand, to a lesser extent, over the low-latitude oceans as well (see discussion above), and these biases occurnearly independent of season. The cold polar bias which is also evident in the zonal means (Figs. 6, 7)can be related to a lack of water vapour in these regions due to the use of the spectral transform methodwhich tends to produce spurious condensation ("spectral rain"), particularly in mountainous areas suchas Antarctica or Greenland.As noted above, the ECHAM3 errors over the low-latitude oceans are significantly reduced as a resultof the enhanced physics. The impact of the increased resolution can be seen particularly in the oceanicstorm track region in the Southern Hemisphere (SH) between about 40°S and 70°S where the warm biasof the T21 versions (Figs. A l and A2, d- f) is larger than in both samples of the T42 simulation (Figs. A land A2, b,c).In ECHAM3, the most persistent error patterns appear over the subtropical oceans in the high pressurecells off the west coasts of North America, South America, Africa and Australia, where the simulatedtemperatures are 2-4K lower than observed. Excessive cloud radiative cooling can be excluded becausethe quasi-permanent stratiform cloud decks in these areas are significantly underestimated in all modelversions (not shown). Since the thermal balance in these areas is achieved essentially by longwave radia-tive cooling and adiabatic warming in sinking air (Boville, 1985), a combination of dynamical and radia-tive factors may play a role.The temperature errors over land are much more seasonally dependent and also model dependent.Furthermore, the error patterns over land are less spatially coherent than over the oceans, and in manycases the temperature errors may be related to the inadequate representation of surface inhomogeneitiessuch as orography, surface roughness, surface albedo etc.. Small shifts in the circulation patterns do con-tribute also as, for example, in Northern Hemisphere (NH) winter over North America and Europe wherethe warm bias of the ECHAM3/T42 model (Figs. Alb,c) can be related to anomalous warm air advectionby a low-level flow which is too zonal over North America and too strong with an anomalous southerlycomponent over Europe (cf. Figs. A7,A1 1). During NH summer, the largest error of the T42 simulation(Figs. A2b,c) is found over the low-latitude continents of the Northern Hemisphere with spatially cohe-rent error patterns of more than +2K and peak values of about +6K over the central part of the UnitedStates. The reason for this model deficiency is not known. The temperature error is, however, consistentwith anomalously small precipitation in those areas (Figs. A20a-c) and is also consistent with reducedcloudiness and soil moisture (not shown).The temperature distribution of the stratosphere evolves quite differently from that of the troposphere,51T DJF ECHAM 1 (C t r l ) -ECMWF(obs )30S60S 30N AON0o’8060S 30S 60N30NECMWF(obs)30N BON60S 30S-40•60■40’3*0'■20105001020SON60S 30S 30Na T DJFi iE 50100- 150- 200„ - 250300’ ----- 400-------- 500700-8501000T DJF ECHAM3(Ct r l -T21) -ECMWF(obs )60S 30S 0 30N 60Nt I I I I I I I I 1 ! I l j l 1 I t I I I I I 1 1 I I~ 50■ 100- 150- 800- 850- 300- 400- 500227008501000. j < I I I I |30N BON30S60SC T DJF ECHAM2(Ctr l ) -ECMWF(obs)80S 30S 0 30N 60N1 t I I > I < t I I I I J I t l I I t 1 I I l ! 1 l i i ! I tf T DJF ECHAM3(Ct r l -T42b) -ECMWF(obs )60S 30S 0 30N 60Ni i i i i I i 1 1 i p- rr. 2 . t 2 J U — 2Wt We T DJF ECHAM3(Ct r l -T42a) -ECMWF(obs)80S 30S 0 30N 60N| I 1 1 I I I I I ! I I I I I ! i | I l501001508=—i i H r60N10001 !60N60S 30N30S 30N 30S60SFigure 6 Latitude-pressure distributions of (a) zonal mean temperature (°C) for the DJFseason according to ECMWF analyses and (b-f) simulation errors of differentversions of the ECHAM model. The lower panel (e,f) shows two 10-year samples ofthe same model (ECHAM3/T42).52b T JJA ECHAMl(Ct r l ) -ECMWF(obs )a T JJA ECMWF(obs)60NI I I I I [ 10 30N30S 30S■ I iBON3DS 30N60S60S 30S 0 30N 60N_j__; _ _I 1 i I I _I | I ! '. ■ ■ I I ' - ' I : I I80 . - - , - / .' w s * ......... ? *• r 100r 70 / •: : : : : : : : : :: : - • -??< : * : : • - ::......... • -10 .................. -0 ------------------15020025030040050070010850100060S 30S 0 30N BONT JJA ECHAM2(Ctr ] ) -ECMWF(obs)60S 30S 0 30N 60NuLai I » iU I- 1 S 1 rr 1 1 1 1 J- L -2 =ToT JJA ECHAM3(Ctr l - T21) - ECMWF(obs)60S 30S 0 30N SONI , J ' , 1, .1—L !_] I 1 I I I 1 ■ .1 I I I . . I I I 1 I J !501001502002503004005002) 00700850100060S 30S 0 30N 60N 60S 30S 0 30N 6DNe T JJA ECHAM3(Ct r l -T42a) -ECMWF(obs) f T JJA ECHAM3(Ct r l -T42b) -ECMWF(obs )BON30S 0 30N30N BON60S 30S 0 60SI lI I I I I I • 1 f-- A 50— £-<- 100r j r j eY f -aP” 1 !- 200X? ' -? - * . * ' - -? ' - . - 250X ’ - 300- 400- 500-700-850f-r1000 t i I r~1 i i i | r0 30NT I I | i60S1 i j i I ! I r30S■ ■ rT I | i i80S 60N30S30N 60NFigure 7 As in Fig. 6 except for the JJA season.53and the essential processes are the absorption of solar radiation by ozone, the emission and absorptionof longwave radiation, and dynamical processes including internal dissipation by breaking gravitywaves, for example. As an example of the skill of the models to simulate the lower stratosphere, we showthe temperature distributions at the 50 hPa level which corresponds to a height of approximately 20 km.During NH winter (Fig. A3), vertically propagating planetary waves excited by orography and land-seacontrast are responsible for a distortion of the cold polar vortex by a seasonal-mean stationary wavedisturbance over the North Pacific. This observed feature (Fig. A3a) is simulated by all models, exceptforECHAMl (Fig. A3b) where the temperature gradients are significantly reduced and the resulting flowis much too weak (not shown). Oppositely, in ECHAM2 (Fig. A3c) .and ECHAM3/T21 (Fig. A3d) themeridional temperature gradient is overestimated by nearly a factor of two, whereas the ECHAM3/T42model (Figs. A3e,f) simulates the correct pattern and amplitude as well. The temperature distributionssimulated by the models are consistent with the modeling of the polar night jet which is too strong in theT21 versions of ECHAM2 and ECHAM3 but correctly reproduced at T42 resolution (cf. Fig. 8).Although the 50 hPa errors in the T42 model are considerably smaller than in all of the T21 versions, theresolution plays probably only an indirect role. The large difference between ECHAM1 (which includesthe parameterization of the gravity wave drag) and ECHAM2 and ECHAM3/T2 1 (which do not) indica-tes that internal dissipation via its influence on the momentum budget and meridional heat transfer is ofmajor importance. Consequently, the improved simulation of the T42 model can most likely be attributedto the inclusion of a well-tuned gravity drag scheme (cf., Table 6 ), whereas in ECHAM1 the effect ofthe gravity wave drag is obviously too strong. In conclusion, even low resolution models seem to needan appropriate stratospheric drag for realistically maintaining the momentum budget (McFarlane, 1987).The stratospheric temperature distribution may be used as a criterion for tuning the scheme.During the NH summer, the meridional temperature gradient is reversed (Fig. A4), and the resulting stra-tospheric easterlies are much less distorted than the westerly flow in winter. The temperature distributionis reasonably well simulated by all of the models. The meridional temperature gradient, however, isgenerally too large, nearly independent of model physics and horizontal resolution. This is true, to someextent, during the SH summer (Fig. A5) as well so that additional processes which are identical in allmodel versions such as the parameterization of radiative transfer or inadequate dynamics resulting frominsufficient vertical resolution, for example, have to be invoked to explain the errors. During SH winter(Fig. A6), the meridional temperature gradient at mid-latitudes is slightly too weak, except in the T42model. However, the cold core over Central Antarctica with observed temperatures less than -85°C istoo warm by 5-10° in all of the model simulations.54WindThe observed and simulated zonal mean zonal wind distributions are shown in Fig. 8 and Fig. 9 for themonths of DJF and JJA, respectively. The impact of an (overestimated) gravity wave drag (GWD) in alow-resolution model is apparent from Fig. 8b which shows the zonal wind as simulated by theECHAM1 model. The polar night jet is only weakly indicated (cf. discussion above and Fig. A3b), butthe NH subtropical jet core is well simulated both in strength and position. The NH westerlies, however,are too weak polewards of about 40°N. In the Southern Hemisphere, the westerlies are generally tooweak which is a common feature of low-resolution GCMs (Boeret al., 1991 ). In ECHAM2 (Fig. 8c) andin ECHAM3/T21 (Fig. 8d) the band of NH westerlies is extended further polewards (which is realistic)but also further equatorwards at higher levels (which is unrealistic). Since both model versions have beenrun without a GWD parameterization, the polar night jet is overestimated (cf. discussion above and Fig.A3c,d). In ECHAM3/T21, the SH subtropical jet is stronger and more realistic than in ECHAM1 andECHAM2. The most realistic mean zonal wind distribution during DJF is simulated by the T42 model(Fig. 8e,f) in both hemispheres and in the troposphere and stratosphere as well. Nevertheless, the modelsimulation is not perfect, and the zonal wind errors are by and large consistent with the temperatureerrors via the thermal wind relation (Fig. 6e,f): The subtropical jets are slightly too strong in both hemi-spheres, and there is also a slight poleward shift of the jet cores, in particular in the Southern Hemisphere,as opposed to the T21 models which tend to shift the SH subtropical jet equatorwards.During JJA (Fig. 9), the differences between ECHAM1 and ECHAM2 are smaller because the GWDeffect on the circulation is small in summer, and the GWD is also much smaller in the Southern Hemi-sphere than in the Northern Hemisphere during the respective winter season. The enhanced physics inECHAM3 produces stronger westerlies in both hemispheres and weaker easterlies in the tropical uppertroposphere and lower stratosphere, nearly independent of resolution. The impact of increased resolutionis, however, evident in the Southern Hemisphere where only the T42 model is able to reproduce theobserved double jet structure. As compared to the ECMWF analyses, the subtropical jets are too strongin the Southern Hemisphere, particularly in the ECHAM3 simulations, and this feature is consistent withthe temperature errors at these levels (cf. Fig. 7).In Figs. 10 and 1 1 the mean meridional wind distributions during DJF and JJA, respectively, are presen-ted for the ECMWF analyses (upper left) and for the various model simulations. The three-cell structureof the mean meridional circulation and its seasonal variation is simulated with some skill by all of themodels, and the differences between the models are smaller than those evident in the simulation of themean zonal flow. In the lower branch of the Hadley cell the simulated inflow is generally larger than inthe ECMWF analyses, while the opposite holds for the upper branch. The simulated height of the uppertropospheric outflow region is systematically too low.The wind systems at the 850 hPa level (Figs. A7 and A8) are generally well simulated. In particular theT42 model is able to capture the strength and position of the observed pattern with remarkable skill (Figs.55ECHAMl(Ct r l )0 30N 60Ni I s i i i 1 I i 1 i i i IU DJF60S 30Sl i t t ! I I I 110560N30N30S60Sa U DJF ECMWF(obs)80S 30S 0 30N BON• 1 1 L- ' 4 1 * 1 uj- 'r1 1 - 1 ‘i.L u ~ 50- 100- 150- 200- 250- 300/ - 400- 5007000501000T60S 60N30N30Su DJF ECHAM2(Ct ' i )60S 30S 0 3uW BONu DJF ECHAM3(Ct r l -T21)60S 30S 0 3ON 60N201510"E 50100- 150- 200- 250- 300400500700106509009501000T60S 30Sr30NI60NECHAM3(Ctr l -T42a)0 30Nu DJF30S! I . l . l . JBONi !80S501001502002503004005007008500009501000I60ST30S 30N BON60S 30S 0 30N 60N60Su DJF30SECHAM3(Ct r l -T42b)0 30N 60N60S 30S 0 30N 60NFigure 8 Latitude-pressure distributions of zonal mean zonal wind (m/s) for the DJFseason according to (a) ECMWF analyses and (b-f) ECHAM model simulations.56u JJA ECMWF(obs) u JJA ECHAMl(Ct r l )BON 60S 3OSX3ON BONX.J 1 -----30N30S80S30N"TBON30NISOS -60S 3OS6DN30Su JJA ECHAM3(Ct r l -T21)60S 30S 0 30N BONCBOSu JJA ECHAM2(Ctrl)3OS 0 30N BON50100150200250300400500700B50ODO050100060ST 130S 3DNIBONECHAM3(Ct r l -T42a)0 30N BONe U JJA60S 30ST30N"TBON80S 30SECHAM3(Ct r ) -T42b)0 30N BONU JJA60S 30S501001502002503004005007008500000501000“T30NT60N60S 30S/ I-f l -r 0 A, / .FI 1 I I I I IBON30N60S 30SFigure 9 As in Fig. 8 except forihe JJA season.57v DJF ECHAMl(Ct r l )v DJF ECMWF(obs)a30N60NI I |60N, |80S 30S30N30S60St I I J—■'"'op50- 100150-200- 250- 300-400- 500-05o0.5TBON30S SON BON80S 30S30N60SECHAM3(Cl r i -T21)0 30N BONI l l i -t, 1 1 I I I I 1 I -1— L .v DJF30Si i i I i60S~ 0\“ 0, \30N 60N60SECHAM2(Ctrl)0 30Ni l l i l i i . l tV DJF80Si I I 1BON, I60S= 50- 100- 150- 200- 250-300-400- 5000.5700•0.50.5-B50DOO- B50100060NT60S 30N30SECHAM3(Ctr l -T42b)0 30NI I I I J I I I I I I I It----- 0.5 tI <7 5 J //.'ECHAM3(Ct r l -T42a )0 SONI i I I J J I I I I 1 i 1■■ v--— - 0.5 /- —] / - 0 - 6V DJF30Sv DJF30Si l l !MN60N60SI50100150200250300400500700- 850<- BOO•j- B50— 1000■-.-0.6 I 0.5’"[■■'■I ' | 160S 30ST T0—A60N30N30N60S 30S 60NFigure 10 Latitude-pressure distributions of zonal mean -meridional wind (m/s) for theDJF season according to (a) ECMWF. analyses and (b-f) ECHAM model simulations.58ECHAMl(Ct r l )0 SON BONt I I t I t ! I | ' I IV JJA30Sill,60S60S 30N 60NECMWF(obs)0 30NI I i i i i I iV JJA30Si I I i i60NI >60S— - 50- 100150- 200\ -250-300-400- 500700650100060N30N60S 30SECHAM3(Ct r l -T21)0 30N BONl i I i i l , i I l i l i i Iv JJA30S60SI IECHAM2(Ctrl)0 30NV JJA30S 60N60S50IDO150200250300400500700650600850100030N 60N60S 30SECHAM3(Ctr l -T42b)0 30N BONI I I I I I 1 ! I I I I I 1 > I I i.7:°v JJA 130Si i i I i ■X -0 .580SI0.560S 30S 30N BON60S 30S 30N 60NECHAM3(Ctr l -T42a)0 30N111 . I I I I I I I I I I I. / 0v JJA30S1 I ! I I t- 0 .560SISON“t 50- 100- 150' -200-250- 300-400- 5007001.5T06508008501000j" 1 I I I I ‘ I60S 30Snr6 ON30NFigure 11 As in Fig. 10 except for the JJA season.59A7b,c and A8b,c). In DJF the tropical easterlies are slightly larger than observed, whereas in JJA thesouthwest monsoon over the Arabian Sea is too weak. In the T21 models, the mid- latitude westerlies inthe Southern Hemisphere are weaker than observed in both seasons (Figs. A7d-f and A8d-f), and thewesterly wind maximum over Eurasia is shifted to the south in winter (Fig. A7d-f). In ECHAM 1, theflow over the North Atlantic is much too weak in winter (Fig. A7d).At 200 hPa, the mid-latitude westerlies simulated by the T42 model are too strong in both hemispheresand and in both seasons (Figs. A9b,c and A 10b, c), as discussed already above (cf. Figs. 8 and 9), whereasthe positions of the jet maxima are well captured. The tropical easterlies are slightly too strong in the T42model during DJF, whereas the tropical easterly jet during JJA over the Indian Ocean is weaker thanobserved. The anticyclonic flow off the west coast of Central America during JJA (Fig. A10b,c) and itscyclonic counterpart at 850 hPa (Fig. A8b,c) are much stronger than observed. The DJF flow at 200 hPasimulated by the T21 models ECHAM2 and ECHAM3 (Figs. A9e,f) is fairly realistic. The westerly biasis slightly reduced as compared to the T42 model, in particular in ECHAM2. In ECHAM 1 , the mid-lati-tude flow during NH winter is generally too zonal (Fig. A9d), and the jet axis over the east coast of theUnited States does not show the observed southwest-northeast orientation. During JJA, the biggestimpact of increased resolution is found in the Southern Hemisphere where the observed southwardextension of the jet in the Indian Ocean sector is only reproduced at T42 resolution (cf. Fig. 9). Theimpact of changed physics (essentially convection) is particularly evident in the tropics where theeasterly jet over the Indian Ocean is significantly reduced in ECHAM3 as compared to ECHAM 1 andECHAM2.Surface pressureThe zonally averaged surface pressure (reduced to mean sea level height) is shown in Fig. 12 for DJF(upper pair) and for JJA (lower pair). There is generally good agreement between the simulated andobserved distributions but a few exceptions are noteworthy. During DJF, the ECHAM1 model simulatesexcessively high pressure over the Arctic (Fig. 12b) which can be explained by the use of a gravity wavedrag parameterization being not properly adjusted for that particular model. Furthermore, all T21 modelversions underestimate the meridional pressure gradient in the "roaring fourties" of the Southern Hemi-sphere (Figs. 12b and 12d): The subtropical highs are slightly too weak whereas the Antarctic trough isnot deep enough. In this respect, the T42 model performs better, particularly during JJA (Fig. 12c), andthis seems to be typical of modem GCMs (Boeret al., 1991).T)uring DJF (Fig. 12a), however, theT42model overestimates the SH meridional pressure gradient, and there is also a southward shift by appro-ximately 5° of the respective pressure patterns. In all ECHAM simulations the equatorial trough is slig-htly too deep.In Fig. 13 and 14 we compare the interannual variability of DJF and JJA means of surface pressure forthe ECHAM3 model simulations at T21 and T42 resolution and for the observations, respectively. The60DJF MSLP - — ANA 3T21 ----- 3T42 ----- 3T421030 -1025 -1020 -1015 -1010 -1005 -1000 -995 -990 -985 -.980 -MSLP mb Q-90 -60 -30 0 30 60 90LatitudeD DJF MSLP - — ANA 3T21 2T21 1T211030102510201015101010051000995990985980J/MSLP mb-30 0LatitudeC JJA MSLP - — ANA 3T21 3T42 ----- 3T421030 -1025 -1020 -1015 -« 1010 -CU 1005 -£ 1000 -2 995 -990 -985 -—179B0_ ■90 -<30J 130 0 30 6 0 9 0d JJA MSLP -—ANA ...... 3T21 ----- 2T21 1T211030 -1025 -1020 -1015 -1010 -1005 -1000 -995 -990 -985 -980 -%/X/MSLP mb9060-30 0 30Lati tude-90 -60Figure 12 Zonal means of mean sea level pressure (MSLP) for the DJF season (a,b)and for the JJA season (c,d). ANA denotes ECMWF analyses, and nTxx denotesECHAMn at Txx horizontal resolution. 3T42 refers to two 10-year samplessimulated with the same model (ECHAM3/T42).61Q DJF 1 MSLP ----- 3T421030 -1025 -1020 -1015 -1010 -1005 -1000 -995 -990 -985 -980 -60-90 -60 -30 30 90MSLP mbLat i tudeb DJF 2 MSLP1030102510201015101010051000995990985980-90 -60MSLP mb----- 3T42- 30 0 30Lat i tude60 90MSLP ---- 3T21C DJF1030102510201015101010051000995990905980-90MSLP mb90"-60 -30 0 30 60Lati tuded DJF MSLP ana1030 -1025 -1020 -1015 -» 1010 -CL, 1005 -1000 -995 -990 -985 -—1 1- 90 -60 -30 0 30 60 90Lat i tudeFigure 13 Interannual variations of zonal mean simulated (a-c) and observed (d) meansea level pressure for the DJF season (10 seasonal samples in each panel).62JJA 1 MSLPa ---- 3T42103010251020 -1015 -1010 -1005 -1000 -995 -990 -985980900 30 60-30-60-90LatitudeJJA 2 MSLP ---- 3T421030 -1025 -1020 -1015 -1010 -1005 -1000 -995 -990 “985 -980 -9060-30 0 30-90 -60LatitudeMSLPJJA 3T211030 -1025 -1020 -1015 -1010 -1005 -1000 -995 -990 -985 -980 -60 90300-30-90 -60Latituded JJA MSLP ----- ANA1030 -1025 -1020 -1015 -1010 -1005 -1000 -995 -990 -985 -980 -90600 30-60 -30-90C/D3c/C «wJa CLI DJF1.0-'-.. . ZZ; - k\.5Q /..@-«Ä°M Sr4•°' 1 '° }V'a - -<= fe1 0-1 .02.0,51.0Figure 15 Comparison of precipitation climatologies (mm/d) for the DJF season accordingto (a) Legates and Willmott (1990) and (b) Jaeger (1976). The lower panel shows thedifference (a) - (b). Contour spacing: .5, 1., 2., 4., 8., ...mm/d.66a2/08.0fiSgfiSiS4j3?o;52.0 — i .o0.5 0.5»ff/*.,., ’ —J7 . \ r<2 .04.0 2.0Figure 16 As in Fig. 15 except for the JJA season.67a DJF prec ip — anaJ - — anaL 3T21 3T42 ----- 3T429030 60060 3090300250200150mm/moq 100500b DJF prec ip 3T21 ----- 2T21 1T21----- anaJ — anaL0-90 -60C JJA prec ipog 250mm/ 200 -150 -u 100 -- 30 0 30 60 90Lati tude----- anaJ - — anaL 3T21 —— 3T42 ----- 3T42-f'- 60 9030-90 -30 60Lati tude300g 250mm/ 200150o wo I ■ * 6RMS(T) DJF60S 80S4— rECHAM2(Ctr])0 30Nj .! I ,i u,. t i I j 1460NI I Irp60N, i l ' I j '80Sunfiltered3QS 30N 60Sunf i l t e r edDJF( 11 -20 ) ECHAM3(Ctrl-T42a)0 3 ON 60N■ 1 ■ 1 1 ! ' 1RMS(T) DJF(21 -30 ) ECHAM3(CLr)-T42b)60S 30S 0 30N GONJ-Au rjk J -Le RMS(T)60S 30SJ—l ! 1 ' h, ! 1 X_L•- 100150J- 200A- 250-300■— 4005002-2-- 700 -ik 850 "!" 900 "I950 H— 1000 “uJj*7 i I I ■?' 1—I—I60S 30Sunf i l te redGOS 30Sc30N GON30N GONFigure 19 Latitude-pressure distributions of zonal mean standard deviation of temperature(K) due to transient eddies computed from the unfiltered time series for the DJF season,(a) ECMWF analyses, (b-f) model simulations. The lower panel (e,f) shows two 10-year samples of ECHAM3/T42.72RMS(T) DJF ECHAMl(Clr l )30S 0 30N BONRMS(T) DJF ECMWF(obs)30S 0 30N BON■ i J i i i i , i , . ! i 1. 1 I t i i j, i L\ / \ \ s >5 ------a60S60S. i . I I0.5 -----1 ■» b.K- 1.5 . X50 =100 -150 -200 -250 -300 -400 -500 -0.6 1.52700 -B60 -900 -050 -1000 -1BON 60S 30Sbandpaaa-filtered03DN30SBOSbandpaaa— fi lteredRMS(T) DJF(2-10) ECHAM3(Ct r l -T21)60S 30S 30N 60bECHAM2(Ctrl)0 30NRMS(T) DJF60S 30Sc60N1 1 d.5 1/ Fs 0.50.6 T3"ÄW.0 .5400-500-700-1 i | n r~l ; J i30N 60N60S 30Sbandpass- f i l teredBON30N60S 30Sbandpaa i - f l l t e redRMS(T) DJF(21 -30 ) ECHAM3(Ct r l -T42b)3DN 60NDJF( 11 -20 ) ECHAM3(Cl r l -T42a)30NRMS(T)BOS60N 30S60S 30SW60S 30S 30N 60N30N 60N30S60Sbondpass- f i l tered30N 60Nbandpass - f i l t e redFigure 20 As in Fig. 19 except for the band-pass filtered frequency range.73RMS(T) DJF ECHAMl(Ctr l )RMS(T) DJF ECMWF(obs)a60S 30S 30N 60N60N30N30S60S50 3100 -150 -200 -260 -300 -400 -500 -i t 2.520.5700 -850 -000 -950 -1000 -*80S 30SlowpaiB-filtared060N 3DN30N 60N3DS60SRMS(T)60S 30SDJF(2 -10 ) ECHAM3(Ctrl -T2 1 )30N 60NECHAM2(CLrl)RMS(T) DJF030N 60N0.5260S 30Slowpass -fil tered0 30N 60N60N30N30S60SDJF(1 1 -20 ) ECHAM3(Cl r l -T42a )0 3 ON 60Nl ! I l I i i l I I I f WTC/A.RMS(T) DJF(21 -30 ) ECHAM3(Ct r l -T42b)60S 30S 0 30N 60NuLl ' I r . l I I, I I I I I I I I I ■ I I I j l I I 11 ' ■e RMS(T)60S 30St I I ! 1. ! ! 1 I: 1 5 ----- \50~ \ X| I 1 I I I J I I I0 30N 60S’| i i l l i | i i 7 i i p0 30N 60N11 1 1 1 P 1 n 1 I60S 30Slowpass-filtered60S 30Slowpast-fi l leredFigure 21 As in Fig. 19 except for the low-pass filtered frequency range.74DJF ECHAMl(Ctrl)0 30N 60Nt I I I I I I I I I 1 I I 1 I30SI n60S50 3"100 -160 -200 -250 -300 -400 - •7008508000501000T*T’ DJF ECMWF(obs)30S 0 30N 60NI ! I ! I I I I I 1 I I J I I I 1 I J_ -a60S30ST60S 60N30N 80S 060S 30N 6CNrJ1 * rp*60SI l l i !DJF(2 -10 ) ECHAM3(Ctrl-T21)30S 0 30N 60NI I I l 1 „1,J I l f l I l j l I I I 1 5 j-400-500-700-i I i i r i i * | i i . >130N 60N' r i i i I | i i i z i i p60S 30ST*T* DJF ECHAMS(Ctrl)30S 0 30N 60N60SD -250-300r- 400\ - 5002700- 650- 900IL 950- 1Ü0060S 30S 0 30N 60!-‘60S 30S 0 30N 60NFigure 22 As in Fig. 19 except for the standard deviation of temperature due to stationaryeddies.75unfiltered data (Fig. 19), for the band-pass filtered data (Fig. 20) and for the low-pass filtered data (Fig.21). In addition, the respective stationary eddy variances are shown in Fig. 22. AU models are able tocapture the observed pattern of the unfiltered and band-pass filtered RMS-values with reasonable skill,but the T42 model performs somewhat better in the tropics and in the Southern Hemisphere. InECHAM1 (upper right panel), the band-pass filtered RMS-values are somewhat too small in the Nort-hern Hemisphere. The low-frequency variability (Fig. 21) is also reasonably well simulated by allmodels in the Northern Hemisphere (except for a lack of stratospheric variance in ECHAM1), but thesimulated tropospheric values are lower than observed in the Southern Hemisphere. The low-frequencyvariability in the tropics is also too small in all models, partly to be attributed to missing interannual SSTfluctuations.In the Northern Hemisphere, the stationary eddy variance (Fig. 22), generated by orography and land-sea contrast, has local maxima close to the surface and in the stratosphere, similar to the transient eddyvariance. In the Southern Hemisphere (summer!), the pattern is quite different with peak values overAntarctica and a secondary maximum generated by the Andes mountains near 30°S. Except for the lackof stationary eddy variance of ECHAM1 in the stratosphere, there is no evidence of a significant modeldependence, and the models agree favourably with observations also.Maps of the transient temperature fluctuations for the Northern Hemisphere winter at 850 hPa and 500hPa, respectively, are shown in Figs. A21-A26. There is generally good agreement between the modelresults and the observations. In particular, the regional separation of the band-pass fluctuations at 850hPa, with large values generated by developing baroclinic wave disturbances over the western parts ofthe mid-latitude oceans (Fig. A22), and peak values of low-frequency variability over the Rocky Moun-tains (Fig. A23), is well simulated by all models. The relatively weak band-pass filtered eddy activity inthe ECHAM1 model (Fig. A22b) is consistent with the poorly developed Icelandic low in that model (cf.Fig. A l lb ) .At 500 hPa (Figs. A24-A26), the impact of increased resolution is evident for the simulation of the band-pass RMS-values (Fig. A25). Especially over the Atlantic Ocean, the T42 model simulates larger andalso more realistic values. Again, ECHAM1 shows the poorest simulation (Fig. A25b) while ECHAM3/T21 (Fig. A25d) is slightly more realistic than ECHAM2 (Fig. A25c) over the Atlantic Ocean whereasthe maximum variance in the Pacific is shifted too far to the east. A similar tendency can also be seen inthe T42 model (Figs. A25e,f). All models slightly underestimate the low-pass RMS-values over NorthAmerica (Fig. A26). The observed low-pass maximum of 4K over Scandinavia is indicated only duringthe first decade of the T42 simulation (Fig. A26e).Geopotential height fluctuationsThe differences between the model simulations on the one side, and between the model simulations andthe observations on the other side, are slightly more pronounced in the geopotential height fluctuations76than in the temperature fluctuations. The poorest simulation is provided by ECHAM 1 with significantlytoo small values in both hemispheres (Figs. 23-26). While ECHAM3/T21 is slightly more realistic thanECHAM2, there is a major improvement with increased resolution in the band-pass range (Figs. 24e,f)where the T42 simulation is close to the observations. The impact of increased horizontal resolution onthe simulation of the low-frequency variability (which is too small in all model versions) is insignificant,however. There is also little evidence of a resolution effect on the simulation of the stationary eddies (Fig.26) but there is a substantial difference between the T21 model simulations, and the largest and also mostrealistic values are simulated by ECHAM3.On the regional scale (Figs. A27-A32), the observed separation of the areas with large band-pass filteredvariance at 1000 and 500 hPa, respectively (Figs. A28,A31), and those with large low-pass filteredvariances (Figs. A29,A32) is well simulated by all models. The impact of increased resolution is apparentin the band-pass range, where the ECHAM3/T42 model simulates larger and generally more realisticvalues than its T21 counterpart. There is, however, little if any improvement in the low-frequency rangeat 1000 hPa (Fig. A29), while at 500 hPa the T42 simulation is more realistic, particularly over Scandi-navia (Fig. A32). The observed level of low-frequency variability is reached, however, by none of themodels.Kinetic energyIn Figs. 27-30 we compare the observed (upper left) and simulated zonal mean transient and stationarykinetic energies. It is evident that all T21 versions are not able to reproduce the observed level of transi-ent kinetic energy (Fig. 27b-d) whereas the T42 model (Figs. 27e,f) simulates the peak values and alsothe distribution with remarkable skill, except for a poleward displacement of the SH maximum. Further-more, the maxima in both hemispheres are displaced upwards to a level of about 250 hPa while theECMWF analyses suggest a height of approximately 300 hPa. The impact of the horizontal resolutionon the simulation of transient kinetic energy is, however, strongly model dependent In a study using theCanadian Climate Center (CCC) model, Boer and Lazare (1988) found only a weak resolution depen-dence between T20 and T40, while Boville (1991) found a systematic increase with resolution (T21,T31, T42, T63) of the transient kinetic energy in the CCM1 model of the National Center for Atmosphe-ric Research (NCAR). Both GCMs suffered, however, from a general lack of transient kinetic energy,and the peak values simulated by CCM/T40 and CCM1/T63 were comparable to those of ECHAM3/T21, for example.The impact of resolution in the ECHAM model is particularly evident in the band- pass range (Fig. 28)whereas the differences in the low-frequency range (Fig. 29) and for the stationary' eddies (Fig. 30) areless obvious. There is, however, a marked impact of model physics on the simulation of the stationarykinetic energy with peak values nearly twice as large in ECHAM3/T21 (Fig. 3()d) as compared to thoseof ECHAM1 orECHAM2 (Figs. 30b, c).77RMS(Z) DJF ECHAMl(Ctr l )30S 30NRMS(Z) DJF ECMWF(obs)30N 60S 60NBON60S 30S6030N 60S60S 30S60N30N30S60Sunfi lteredunfi l teredDJF(2-10) ECHAM3(Ct r ) -T21)0 30N 60NRMS(Z) DJF60S 30S111 J i I I 1 1 1ECHAMZ(Ctrl)0 30N BON\2060N60Sunf i l l e red30S10 30N 60N- 500-7001 ' ' I ’30S60Sun f i l teredRMS(Z) DJF(21 -30 ) ECHAM3(Ct r l -T42b)e RMS(Z) DJF( 11 -20 ) ECHAM3(Cl r l -60S 30S 30N 60N60N60S 30S 30N0\0<\J$•20 Bfc 501 00- 150- 2001 -250i- 300> -400161500!700I I \ \s. i t §i- 650- 900- 950- 100060S 30S 30N60S 30S 30N 60NFigure 23 Latitude -pressure distributions of zonal mean standard deviation of geopotentialheight (m) due to transient eddies computed from the unfiltered time series for the DJFseason, a) ECMWF analyses, (b-f) model simulations. The lower panel (e,f) shows two10-year samples of ECHAM3/T42.78RMS(Z) DJF ECHAMl(CLrl)30S 0 30N 60N>.J .1 I H I I I dRMS(Z) DJF ECMWF(obs)60S 30S 0 30N 60NI 1 ! L > 1 I l l ■ i i I l l i l I I I i I J I I60SSOL 30202 §TSON1000 » i I I I I i i i T I60S 3OSbandpaaa- filtered60N7'1 I I I | 1 i i i I |60S 30Sbandpaas- filtered30N 6 ONRMS(Z) DJF(2-10) ECHAM3(Ctrl-T21)60S 30S 0 30N 60Nf t t I 1 I I I I I I I I I I I I I I I I I I I I I I I I I I • < t •— ■■ -L - ceRMS(Z) DJF60S 30Si 1 ! ! I I I ! . J O I 1ECHAM2(Ctrl)0 30N 60NI I 1 ! I > I I I f I t I I302030' ' 1 ! 1 | .................. | i i i i i | 1 , i i i | i 1 r '[' > > i < •60S 30S 0 3 ON 6 ONbandpass -fi ltered60S 30Sbandpass-f i l te red0 30N 60NDJF( 11 -20 ) ECHAM3(Ctrl-T42a)0 30N 6ONe RMS(Z)60S 30SRMS(Z) DJF(21 -30 ) ECHAM3(Ctrl-T42b)60S 30S 0 30N 60S10001 r ■ - 1 • i30N 6ON. i ] i i i r r p60S 30Sbandpass-filteredi < ; < i | < i I I I I60S 30Sbandpass-filteredn ■ ■ 1 1 1 1SON 6 ONFigure 24 As in Fig. 23 except for the band-pass filtered frequency range.79RMS(Z) DJF ECHAMl(Clrl)Q RMS(Z) DJF ECMWF(obs)60SI30Sdj30N60N 60N30N30S60S608 ?30Nnr30NI30S 60N 30S G ON60Siowpaea -f i l teredd RMS(Z)60S 30SDJF(2 -10 ) ECHAM3(Ctr) -T2 1 )0 30N 60NECHAM2(CLrl)C RMS(Z) DJF60S 30StOO806 ö oc\> xr co0 30N 60N8 ? §i i i | i i i i i I i3ON 60N 60S 30Slowpass— filtered60S 80S 0lowpeia -f i l teredDJF( 11 -20 ) ECHAM3(Ctrl-T42B) f0 30N 60NRMS(Z)60S 30SDJF(21-30 ) ECHAM3(Ctrl-T42b)0 30N 60 N1 I 1 ! I I I 1 i I I I I 1 > I I ie RMS(Z)60S 30SI 1 t l.nl I '■ I 1 I t I !ec-1000 I I I I l | I I I I l |60S 30SlowpaM-fllteredI | I I I I i |30N 60Ni I I I I I I H I I | I60S 30Slowpass - f i l tered30N 60NFigure 25 As in Fig. 23 except for the low-pass filtered frequency range.80Z‘Z* DJF ECHAMl(ClrJ)0 oOi'i uüpI I I I ' I I I t I t I I '60S 30S50100150200250300400 -500700850800nr30N3QS 60NDJF ECMWF(obs)0 30N BONI i ! ■ i ; 1 i j . 1 ia60SZ-Z-30S3DN 60N30S80SECHAM2(Ctrl)0 30NDJF30S<0.z*z6ON60S4060 40 -60'-8O--DJF(2 -10 )30S 0ECHAM3(CtrJ -T2 1 )30N 6 ONZ*Z* DJF(11 -2O) ECHAM3(Clrl-T42a)30N BONl I ............ i ;60S 30SB 50rj- 100f r i so1 -2001 -250i -300: i - 400r- 500-SO. z 60 .40 700B509009501000Z*Z* DJF(21 -30 ) ECHAM3(Ctrl-T42b)60S 30S 0 30N BON■ ' i , t t i I i < < i t ! i i t i i I i ' > * • • i . .BOS 30S 30N BON-60S 30S 30N BONFigure 26 As in Fig. 23 except for the standard deviation of geopotential height due to *stationary eddies.81DJF ECHAMl(Clrl)0 30NI I I l I I I 1 I I I I fKTE60S, I30S50100150200250300400&050040700850800950100030NT i ' I I I | I60Sunfi ltered6ON30SO KTE DJF ECMWF(obs)60S 30S 0 30N 60Ntirrrrv r. 60200.16C60*060S 3DS 0 30N 60NunfilteredDJF(2 -10 ) ECHAM3(Clrl -T2 1 )30S 0 30K 60Nt I I I i i r I i i i 1 i i t • i I60<-ISO'60 ' 80ÖCKTE DJF30SECHAMS(Ctrl)0 30NOOS aoN120JOD- äO«0'S ___ __ yI I I I I I I I ISON 60N40I 1 I I I I I I60Sunf i l leredT30S 30S 0 30N 6ON60Sunf i l t e r edf KTE DJF(21 -30 ) ECHAM3(CLrl-T42b)e60S■ ' > • ‘ iKTE DJF( l l -20 ) ECHAM3(CLrl-T42a)30S 0 30N 60Ni I ! I I I I .1 I T I I I I I I | I 1 I I I I ■"HP ■ 50- l oo___r- 150— 2002501- 3 ”0lütt' ßC - 500<■----- - 700/ 1 i/ ' !. - 050( - 900x- 950----- ----------- 1000l{‘Ä60S 30S 0 30N 60N£206$:120"SO"26tF>24100"60S 30S 0 30N 6CN3CX 60SunfilLeredFigure 27 Latitude-pressure distributions of zonal mean transient eddy kinetic energy (m 2 /s 2 ) computed from the unfiltered time series for the DJF season, a) ECMWF analyses,(b-f) model simulations. The lower panel (e,f) shows two 10-year samples of ECHAM3/T42.82DJF ECHAMl(Ctr l )0 30N 60NI I [ I I ! I 1 I ! I I I ■ IKTE30Si I 1KTE DJF ECMWF(obs)80S 30S 0 30N BONI - : 2 I J 1 ,1 I, I I , I I J I I ! I ! 1 I 1 I I J. j p l I 1 !■_a60Ssog-100 -I160 -00 -250 -300 -400 -600 -103060202040 2020 7008600000501000gI 1 I I I I I I I60Sbandpaas-filterede>q-r30Nrp60N60N 30N30S30S60Sbandpaci - f i l teredDJF(2 -10 ) ECHAM3(Ct r l -T21)30N 60NKTEECHAMS(CLrl)KTE60SDJF90S 60S(Q ,1 1 I 1 I l 130S 060N30N305030400-102020700-60S 30Sbandpass - f i l tered0 30N 60N30S BON30N60Sbandpaca- f i l teredKTE DJF(21— 30 ) ECHAM3(Ct r l -T42b)80S 30S 0 30N 6ONDJF( l l -20 )30S 0ECHAM3(Ct r l -T42a )30N 60NKTE60Se= 50- 100k 150L- 200q -2501 -300-400’ - 500*60’■40 J'2<' 20 ' ‘2170060Sbandpass - f i l t e red30S 0 60Sbandpass - f i l t e red30S 30N 60S30N 60NFigure 28 As in Fig. 27 except for the band-pass filtered frequency range.83KTE DJF ECHAMl(CLrl)KTE DJF ECMWF(obs)a60S 30S 01 I I I I I I I30N 6ü.‘.50 h100 -160 -200 -250300 - (400 -204060020700850 -800 -060 -1000 -* i i r~i r 1 T r i j j i i , i60N60SIowpass-filtered30S 30N60N3ON030S60S60aoBO4040< i | I I I I I | l I0 30N BON30S80Slowpass -f i lteredKTE DJF(2 -10 ) ECHAM3(Ctrl-T21)60S 30S 0 30N 60Ni i 1 i t t i i I t > i i i I t i i t i I i i i t ■ I i t t > 1 IKSE30SI ■ ,60SI50 —100 -150 -200 -250 -300 -400 -500 -20700asoooo9501000T30S“T30N60S 60NKSE DJF ECMWF(obs)30S 0 30K 80NI I i J I J_I I 1 I I U.20 . y y'a80SIi 1 1 ' ' 1 | ' ' 1 I 1 | 10 30N eONI 1 I I f I I80S 30SDJF(2 -10 ) ECHAM3(Ctr ] -T21)30S 0 30N 60NI I T I I I I I J I 1 I 1 I I ! I I < ! ■ • I fC KSE DJF60S 30SECHAM2(Ctrl)0 30N 60N124*700-60S 30ST30N 0 30N 60NAON30S60Se KSE DJF( l l -20 ) ECHAM3(Cl r l -T42a )60S 30S 0 30N 60NI t i 1 I I i i I I I I l l l i i I i i > 1 i I I I 1 i I 1KSE DJF(21 -30 ) ECHAM3(Ct r l -T42b)60S 30S 0 30N 60N........ "50100150200250300400500Q)2?■80'- 0C■40700B509009501000■|' i i i r i | i i i i i | i0 30N 60NI TT TTT "FT“60S 30S| I I I i I | I I I I I |0 30N 60Ni I I I i I | •60S 30SFigure 30 As in Fig. 27 except for the stationary eddy kinetic energy.85All of the differences discussed above are evident on the regional scale as well (Figs. A33-A35). Theunfiltered transient eddy kinetic energy at 300 hPa (Fig. A33) is significantly higher at T42 resolutionthan at T21 resolution and also more realistic although the simulated values are systematically smallerthan the observations suggest (Fig. A33a). As discussed above, this deficit does not indicate a generalunderestimation of the transient kinetic energy in the T42 model but rather an upward displacement ofthe level where the peak values are simulated (cf. Figs. 27a and 27e,f).The impact of increased resolution is significant in the band-pass range (Fig. A34) while the impact ofchanged physics is noticeable only between ECHAM1 and ECHAM2 in the North Atlantic area. In thelow-frequency range (Fig. A35), all models fail to simulate the observed values. Nevertheless, the simu-lations seem to gradually approach the observations with more elaborate physics (ECHAM1 toECHAM3) and with higher resolution as well.Momentum fluxesOne of the most serious deficiencies of the ECHAM0/T21 model is a drastic underestimation of the meri-dional fluxes of westerly momentum by transient eddies (Ponateret al., 1990). To some extent, this short-coming is also apparent in the more recent low-resolution ECHAM versions (Figs. 31-33) although thesimulations are much more realistic in the Southern Hemisphere as compared to ECHAMO. In the band-pass range, there is also a gradual improvement from ECHAM 1 to ECHAM3, particularly in the Nort-hern Hemisphere (Figs. 32b-d). The most dramatic change is caused, however, by the increased horizon-tal resolution in the T42 model with nearly a doubling of the band-pass filtered westerly momentumfluxes in both hemispheres. In fact, the simulated values are generally larger than the ECMWF analysessuggest.The impact of increased resolution can also be seen in the low-pass range (Fig. 33) where the T42 modelsimulates significantly larger momentum fluxes than the T21 models but still underestimates the obser-ved fluxes by typically 30%. In this frequency range the impact of changed physics is hardly detectable.The stationary eddy momentum fluxes (Fig. 34) are negligible in the Southern Hemisphere (during thesummer months!), but they have a distinct maximum of northward transport in the Northern Hemisphere(during winter!) associated with a southwest-northeast tilt of the stationary troughs and ridges betweenabout 15°N and55°N (cf., the time-mean 500 hPa height distribution shown in Fig. A15). The southwardtransport of westerly momentum, on the other hand, indicates a northwest-southeast tilt of the respectivetrough and ridge axes which is unrealistically strong in the ECHAM 1 model (cf. Fig. A 15b) but notstrong enough in the other model versions. Apart from the deficient ECHAM 1 simulation, the modelsimulations are fairly realistic, and there is less model dependence than for the transient momentum flu-xes.Maps of the transient eddy momentum fluxes at 300 hPa (Figs. A36-A38) indicate that, apart from the86ECHAM1 model, all models are able to reproduce the observed large-scale pattern of the unfiltered flu-xes (Fig. A36), such as the large pole ward transports of westerly momentum over North America, NorthAfrica and East Asia, and the southward transports to the north of about 50° N with peak values over theEast Pacific and over the North Atlantic. Over North America and over the East Pacific, all models overe-stimate the northward transport of westerly momentum, particularly at T42 resolution. The observedlarge poleward fluxes over the Central Pacific are not reproduced by any of the models.In the band-pass range (Fig. A37), the largest model errors are found over the Pacific and over NorthAmerica. The ECMWF analyses show a distinct dipole pattern over the Central Pacific and a clear sepa-ration of this pattern from an area with large northward transport of westerly momentum over NorthAmerica (Fig. A37a). In the models, however, except for ECHAM2 and ECHAM3/T42 during the firstdecade (Figs. A37c,e), the Pacific dipole is but weakly developed and the maximum over North Americais too pronounced in the T42 simulation. In ECHAM1 (Fig. A37b), the areas with large southward trans-port are too extended while the northward transport is much too weak. In the Atlantic-European area, themost realistic simulation is provided by the T42 model which is able to reproduce the characteristic sout-hwest-northeast orientation of the dipole pattern with remarkable skill (Figs. A37e,f). The increased con-vergence of the momentum flux over Scandinavia and England during the second decade of the T42simulation is consistent with the deepening of the Icelandic low and with the enhanced flow in the wholetroposphere over that area (cf. Figs. A7b,c; A9b,c; Alle , f ; A15e,f).The notorious underestimation of the low-frequency variability in all ECHAM simulations is also appa-rent in the respective momentum fluxes, particularly in the low-resolution models (Fig. A38).Heat fluxesThe zonal mean meridional heat fluxes by transient and stationary eddies (Figs. 35-38) are well reprodu-ced by the models. The largest model errors are found in the upper troposphere, particularly in the T42simulation, where the simulated heat fluxes are significantly too large, consistent with the overestimatedmeridional temperature gradient at these levels (cf. Fig. 6).In the band-pass range (Fig. 36), all models perform reasonably well in the lower troposphere (exceptfor ECHAM 1 in the Northern Hemisphere), while the upper tropospheric meridional heat fluxes areagain too large in the T42 model. The deficit of low-frequency variance is less obvious for the heat fluxes(Fig. 37), and the model dependence of the simulated heat fluxes is not very pronounced. This appliesalso to the simulated stationary eddy heat fluxes which are reasonably well simulated by all of themodels, except for ECHAM 1 in the stratosphere (Fig. 38) where the simulated values are much toosmall.The maps of the meridional heat fluxes by transient eddies at 850 hPa (Figs. A39-A41) show a distinctregional separation of the band-pass filtered fluxes with peak values generated by developing baroclinic87u'v ' DJF ECHAMi(Clrl)30S 0 30N 60NI I I « I I I I I I 1 I ! < i i i i60Sd-L-i !5010015020025030040050020107008509009501000 Qi0T30S 30N 60N60SunfilteredV DJF ECMWF(obs)30S 0 30N 60NI 130S| I 1 » I I | t i I l ’ |0 30N 60N80Sunfilteredu 'v '60SDJF(2 -10 ) ECHAM3(Ctrl-T21 )30S 0 30N 60 \I < t i * i - i i < ' . 0 30N 60N30SI i ; I I [ I I I I i | -60S 30Sunfillered60Sunf i l te red0 30N 60Nu 'v ' DJF(21 -30 ) ECHAM3(Ctrl-T42b)30S 0 30N 60N1-J. I I I : t i -1 I J 1 * ■ I * * , 1 ' 1 'l.60S230N 60N60S 30SunNLcredu 'v ' DJF( l l -20 ) ECHAM3(CLrl-T42a)60S 30S 0 30N 60N; I | ■ I ; I 1 < ! < j t I I I < I I I ■ ! I 1 L . ' ;— 50100I- 150i-SOO1-250t- 300e1 -400f I- 500I ii 0- 700-650- 900-950— 1 00 "60S 30S 0 30N 6GXunf i l te redFigure 31 Latitude-pressure distributions of the zonal mean northward transport ofwesterly momentum by transient eddies (m 2 /s 2 ) computed from the unfiltered timeseries for the DJF season, a) ECMWF analyses, (b-f) model simulations. The lowerpanel (e,f) shows two 10-year samples of ECHAM3/T42.88u'v' DJF ECHAMl(Ctrl)0 30Ni ■. | _ | i i I i i i60SI !30SI60N50-100 -150 -2 1 -250 -300 -2400700850B0085010Q0eosbandpass-filtered30S 30N 60NuV DJF ECMWF(obs)60S 30S 0 30N 60NI i i I I t I I i > i i i i ■ i i i I I i i I ■ I i i io \ 'S.« 760NI30N• i i I | I I60Sbandpass-filteredT30Su'v' DJF(2-60S 30S 0I) ECHAM3(Ctrl-T2 1 )30N 6üNit > » I i i i i ; « » t r »Z ------- CHO'30N 60N60S 30S 0bandpass— f i l tered60N30N30S60Sbandpaaa-fil LeredECHAM2(CLrl)0 30NC u 'v ' DJF60S 30S SON2u’v ' DJF(21 — 30 ) ECHAM3(Ctrl-T42b)30S 0 30N 60Nu’v ' DJF( 11 -20 ) ECHAM3(Ctrl-T42a)30S 0 30N 60NJmt i l60S60SI I l i1=_______A \ —2'M . ' pm > p60S 30Sbandpass-filtered60S3 ON60S 30Sbandpass - filtered30NFigure 32 As in Fig. 31 except for the band-pass filtered frequency range.89DJF ECHAMl(Ctrl)0 SON 60NI i I I ( 1 1 i I I I 1 ■ . 'DJF ECMWF(obs)0 30N SON-1 » 1 » 1 U ■ i j 1u'v’60S 30SI i t i i i I60S501000"T" 1 1 i [ 160Slowpan-filtered7~1 l i i | F l60Slowpass- filteredT30Sr30Sr60N30N30N 60NDJF(2 -10 ) ECHAM3(Ctr) -T2 1 )30S 0 30N 60b.u ' v '60SECHAMS(Ctrl)0 30Nu’v’ DJF80S80SI I60Ni i r i i | 4 i60Slowpasa— filteredT30S 30NI I I I i | i | | | | | |60S 30Slowpass -filtered60N060N 30Nu’v’ DJF(21 -30 ) ECHAM3(Ctrl-T42b)60S 30S 0 30N 6ONt i l l I L I I 1 I 1 t 1 t I I 1 t 1 ! C : -DJF( l l -20 ) ECHAM3(CLrl-T42a)30S 0 30N 60N! I 1 1 1 1 1 I 1 1 t 1 > * 1 I ■ 1- ! — 1 1 >e u 'v '60S1 1 1 1 1 I 1 1 1 1 50S- I 501-200: i - 250r- 300-400-500-700C-B50-900-9501000 I | l i 1 I I |30N 60Nr I I I I | 1 1 i 1 1 |60S 30Slowpass- filteredf t 1 r r | ! 1 1 i 1 |60S 30Siowpass- filtered30N 60NFigure 33 As in Fig. 3 1 except for the low-pass filtered frequency range.90DJF ECHAMl(Clrl)0 30N 60 .Ni i I i i i i i I i i i 1 i Lr - 5U ’V ’30Sd-LJ.60S50IOC150200250300400500700850SOO950100060S~r30S 60N30Nu'v’ DJF ECMWF(obs)80S 30S 0 30N 60b' ’ L, 1 1 ' ■ ■ ' ' ' 1 fa ’ • ' _L 1 ‘ 1 ’ AcL.ECHAM2(Ctrl)0 30N 60Nu*v* DJF60S 30SI I 1 I 1 I I I I IECHAM3(Clrl-T2 1 )30N 60NjJj 1 < .Lil» 6 : o . ! 1♦00 30N 60NU*v* DJF(21 -30 ) ECHAM3(Ctrl-T42b)60S 30S 0 30N 60N■ ■ 1 l i s I < ; i I I I30N 60N60S 30SUV DJF( 11 -20 ) ECHAM3(Clrl-T42a)3OS 0 30N 60N> i > i I « i i 1 1 i 1 l _ | ijie60S50100150200250300-400i r 500■20-10 'i ■ °M—60S 30S-700i- B50| ,.— 9001 ’ L P5Q----- ) 0003CX 60NFigure 34 As in Fig. 31 except for the northward transport of westerly momentum bystationary eddies.91vT DJF ECMWF(obs)60S 3DS 0 30N 80N‘ I I 1 ' : I I I 1 ! I 1 I I ! > I I I L J . ! . l ! . ,a vT DJF ECHAMl(Ctrl)60S 30S 0 30N CON■W I I I 1 I I I ! H I I 1 I 1..I.50100150200250300400500210700850900950 I- ”1 1 I30SI30S 60Sun f i ltered30N60Sunfi l tered6 ONsonDJF(2-10) ECHAM3(Ctrl -T2 1 )30S 0 30N 60NECHAM2(Ctrl)0 30NDJF30Sc60S■ I , , I I60NI 1 I I t i00I I I I I I60Sunfi l tered60Sun f i l t e red30S 030S 30N 60N 30N 60NvT DJF( l l -20 ) ECHAM3(Ctrl-T42a)305 0 30N 6ON=- =50") ) 10°_r~\ > > / r 150is-/?: • : • > -------- ) //U((C. I- 200a- 250<- 300■—400500700650900950100060S 60Sf i l t e r t evT DJF(21 -30 ) ECHAM3(Ctrl-T42b)30S 0 30N 60NL f ’ 1 Lu-'o* ! 1 k ; 1 1 J_Lk->— c , , '60S1 / 2 v/.n.-rgFigure 35 Latitude-pressure distributions of the zonal mean northward transport of sensibleheat by transient eddies (K m/s) computed from the unfiltered time series for the DJFseason, a) ECMWF analyses, (b-f) model simulations. The lower panel (e,f) shows two10-year samples of ECHA? T42.92vT DJF ECMWF(obs)60S 30S 0 3ON BONI ! t I » I I I > I I I I I > ■ I I 1 .1 1 i 1 < IvT DJF ECHAMl(Ctrl)30S 0 30N 60NI l l i 1 1 I i i ) i i I i i i i ■60 -100 -150 -200 -250 -300 -400 -500 -’ ' ■I • t * *! < • / \ \• • • • / * \ «2700B509009501000o0H-! I I I l ' t l • - < |60S 30Sbandpasi- f l l leredTBON 60Sbandpaaa-fUlered303 30N BON30NECHAM2(CLrl)30N GONi |j_i i i i_jvT DJF60S 30SI q IvT DJF(2 -10 ) ECHAM3(Ctrl -T2 1 )60S 30S 0 30N 60N/ J " 3// / // / AJ;• i i \W5 I /.-&400500-700-10'60S 30S 0 30N 60N60S 30Sbandpass- f i l tered30N BONbandpass - filteredvT DJF(21 -30 ) ECHAM3(Ctrl-T42b)G vT DJF( 11 -20 ) ECHAM3(Clrl-T42a)60S 30S 0 30N 60N 60S 30S 0 30N 60N50100150200250300400500700B50900950100060S 30S 0 30N 60Nbandpass- f i l l e red: i r | . ’ i - . . | i - : i —| T ► ; i i |’ i“ • "i "i - “ i60S 30S 0 30N 60Nbandpass - f i l l e redFigure 36 As in Fig. 35 except for the band-pass filtered frequency range.93ba v’T' DJF ECMWF(obs)60S 30S 0 SON 80NJ I t I I 1 I I -L I I j j * 1 * * J I I L° •T “ f TeoNT30S 30NECHAM2(Clrl)0 30NI i i i j i I ■ i60S 30Slowpass-filteredi I ff rrTf /30N 60NTov 'T ' DJF(21 -30 ) ECHAM3(Ctrl-T42b)60S 30S 0 30N 60Nfa i > t i 1 t t t t > I t t t i i L i » « « i I i i i ■ 1. 1 ■ 1 r- A«.t i tDJF( 11 -20 ) ECHAM3(Ctrl-T42a)30S 0 30N 60Nl I t I > I l J 1 I I I l I ■ I I » l ■ T-U. I I I■ 0e vT'60Sj ';Vi J 1 1I cF50100150'o 'JJ ~" g /I i I I I | I I I I l | I60S 30Slowpass- f i l te red1 5 i r/ i0 3DN 60Ni | 1 - » i I | 130N 60N60S 303lowpass-f i l teredFigure 37 As in Fig. 35 except for the low-pass filtered frequency range.94v*T* DJF ECMWF(obs) b V*T* DJF ECHAMl(Ctrl)60S 30S 0 30N bC;.80S 30S 0 30N BON50 T100 -150 -200 - =260 -31 I - >400 -600 -700B50BOOB60100060S 30S 0 30N 60NE+■turnI * z> ■T"l . I | I I I I 1 |60S 30S 0 30N BONv*T' DJF ECHAM2(Ctrl)30S 0 30NV*T* DJF(2~10) ECHAM3(Ctrl -T2 1 )60S 30S 0 30N 60N■ I . . . . . I i i * i I > Jc60S BONfew400 - , . \500- •>, °700-1;;;- 10i j i j • i q- i V i i30N 60N1 1 1 I i | 1 1 > 1 1 | '60S 30S60S 30S 30N BONf v*T* DJF(21 -30 ) ECHAM3(Ctrl-T42b) -6 v*T* DJF(11— 20) ECHAM3(Clrl-T42a)60S 30S 0 30N BON60Si I30S 30N 60N, I>fQ /Mfe1T21 DJF&— ;-i-. __2T21 DJF0 id?'• - 1 •;- ---- ---:< 2° .. .Al------ 5 ’.• F' -42....."/ : S- J2 ' ''W XSF’<...-1 .-.? _”A . .........■Ädzzi r? - <-.----- 1-ilFigure A 1 (cont.)113" ■15 --------P1K ; 15 ! —3T42 1 JJAFigure A 2 As in Fig. A l except for the JJA season.1147, 17. ...... .7 Ä1T21 J JAeI I•;/ (2b ’Vc»Ätl’ .7?11;. " ~ 1 ' 'sslT n s— 1:::: : 2 li2~s«ta*®2T21 J JAZ™. LtT tttS23T21 J JAFigure A 2 (cont.)115655 T J / /'M/ Z '/ 7° 7-j65CLIM DJF 65'-65_y /65 /-X Ö— 65—L------ ?'6565-,65 7®Ö--2T21" DJF-Xr-“fF ,7065.65'65-65 ' '3T21 DJFe65Figure A3 a) Climatological distribution of temperature (°C) at 50 hPa in the NorthernHemisphere for the DJF season based on ECMWF analyses. b)-f) Simulateddistributions obtained with different versions of the ECHAM model. For modelnotations see Fig. Al.116a60“X -60 ,' k TOfW- x».65r-05-1T21 JJA -65-6060\ ’’\'' f?\fe\ Ss1 \ \51516<602T21 JJA6'3T21 JJA6i/ Xi Vi?3T42 2 JJAFigure A4 As in Fig. A3 except for the JJA season.117Ä/■■ - -504b-J /— -60’50-45——50’65-65 -CLIM DJF 1T21 DJF. . . . j .. ...65 — l 65-6045 -50: 515-45;.50 J.■' -6065"djfT65 riAT 502T21 DJFsß-\ ( i 1"M y -----e-£5A-45 \45 .d—• \w ■ V.WO;, ------ +r65 65( \ V\ r ; z ' -65T 50 3T42 2; AZ,/ /a 60 a \ ?\A 65 -- M; T 50Figure A 5 As in Fig. A3 except for the Southern Hemisphere.118i60'65\-60$iO/ J6060 S\/i if X r60CLIM JJA 6o__\ l ox \ _ 6(r— po-s-lXx t ieoX... r"W >T 50612T21 J JA~ 3 -60 JJAr, \ -60"¥60i0-6O.__3T42 1 J JAFigure A 6 As in Fig. A4 except for the Southern Hemisphere.119oo,cef»S-qfnvir '-.ÄS**—ty - :•■•• s tFigure A 7 a) Climatological distribution of wind (m/s) at 850 hPa for theDJF season basedon ECMWF analyses. b)-f) Simulated distributions obtained with different versions ofthe ECHAM model. For model notations see Fig. A l . Contour spacing: 5 m/s. Stipplingindicates areas with V > 10 m/s.120pz 3- >.■> >->!*■-~*-~z—*~ ‘ --* s %. j *1T21 DJF"ri "?4 y2t1T DJF ~(K'S3T21 ' DJF "Figure A 7 (cont.)121acuJJACLIMbx 3*"“*VBSQ. l 0.8*Figure A 8 As in Fig. A7 except for the JJA season.122:...■X 7 7 7 '•'■ ‘ - ' ’ 7' ■••••!•<■&■-$"- - -x s i-offe, ' i% - i - - -l » . - | 4 ' i ' ' -Ja-c r~7 ~0:2 :, yv >?»"%•? :<7. 30 m/s.1241— - =-10-;— ---- 71T21 DJF i?- -~.:. iV 200:löijiOL,. _w oou s < -:V 200:Figure A 9 (cont.)125a> i i tL I -Ö.2"' pw s':-:x.r j-.’ :- ■Ui'-A- — lo 7. j-Vj"" " " ’ ’ ~ ~' ''■» i » **y 1 ’ jiffUwjWi ■;' <■> jj ‘fi- ' t-'Mjpt-’15*-i-’ •~ i-ao-N: ■p£q:Figure A 10 As in Fig. A9 except for the JJA season.126■W Z“zzzw 1AAAsS2EEIZ zZ.. zAA±V 2001T21 JJAel-Q: ; > s«-x..: ■: . „ jvA. - 2 1A ; A ~ *_ZEZA-» i"**•■•> — iJrA.iii?.....!!!...?:itei;x>iHijt»ij fWIXJEHtT72;>v••:ä V-älTC ü ibi i H>i ti ii»> >i JäOstV 20032T21 JJAezzzazzz rsaS MEfzT?: p : -t' f2•T. ! - - • ■ c....r-...,A dAizA.ZA—* - i —► —* —» -> !-*-=» —>— *PQioAo- A4 cV / \\iv"‘200Figure A 10 (cont.)12710151015 ■- -------/Z 5~io iot-1015 101 P 1Ö10izz: ioio~ \ /1.0 1'0< \1015 f/'\ 1015'rf'öao • '10151020 "« v<\ \ i p i :•/Z-löIbAsr-T' r\j o t bi' 1015\Z1020 „zT’ OCLIM DJF- 10*r 1020C MSLP\ 10151T21 DJF"'’ 10!20-1015 10101015 10K5\1,01010 15102010203T21 DJFV 1020 ■*,2T21 DJF MSLPe •'1010z 1015•1020-1 .1020. 10 Iß’1010'ipro.r 02v! \K010n*°XJJMSLP2T21 DJF.1010e10101015TOi1015 z - \101p ' ' 101 §.Figure A 13 As in Fig. A l l except for the Southern Hemisphere.1301020_! __Ai•777-1015."pasZTSSX,1015’MSLPa Mö (3020•- . MSLP■Pi Z th 15 ' '■10202T21 J JA MSLPd■ 1020 /• i \1020- —Z 10153T21 JJA MSLPMSLPU- 020 1025— d 102Ö\ dl— iV -lojao—1020MSLPFigure A 14 As in Fig. A12 except for the Southern Hemisphere.131584—....576.i \ \560CLIM DJF r 3 584"Z 500 1T21 DJFJ76-458484'58450 Z 500 3T21 DJF-s2T21 DJF576 -----?P 4. 68 ==$:.04.‘5.’584Z 500Figure A 15 a) Climatological distribution of 500 hPa geopotential height (dm) in theNorthern Hemisphere for DJF based on ECMWF analyses. b)-f) Simulated distributionsobtained with different versions of the ECHAM model. For model notations see Fig. A l .1325B4-0< LZ 500.584'2T21 JJA Z 500;e:584’ XZ 5003T21 JJA~ Te592,584~.68/ 73T42 2 JJA i7 \ i.Z 500■59:a,584 z591Z 5007—7<-<~S9;1 JJA iFigure A 16 As in Fig. A 15 except for the JJA season.133a i I— 584 .-'-"C-576— <•576 — ---2584DJF__ 5$4 -----4>5765 4 -5j4-°4 ö ' \\\\\\5844...... \ 536 \ \ \ \ \4> 5365l 584..........■'■-'"i,.- 1:-.’584.... >Z 5003T21 DJF :i-----5ffI584 .... r"W >Z 500Hlit]//öß4■\ j. rV ,Z 500584•5l3T42 1 DJF iFigure Z 17 As in Fig, A15 except for the Southern Hemisphere.134.584-5 68.52i12576,560'5B45761T21 JJA584 :576 =584-5 Z 500■568— -1'504\_584.576' «QJ - -2T21 JJA 576 ' Z 500.584 — j~~58|.45 568=4?584564568-U--|—576576'5003T21584,zz \>Axzzt_A- 57B5< A~' , ’3T42 1 JJA— 5 ?4 "Z 500-s59€ ! _''' \ .Re-A1 ?5B>4 \ \ v\ \ Ä56i584t ■■ ■>/ 1Z 500eFigure A 18 As in Fig. Al 6 except for the Southern Hemisphere.135«SSsHsfeKo;$2.0W2äaa®2?42.8 ’ :'------------- -------- TK4.o_ ;:2,0CLIM DJFX ’ rifrh .= • ! ! • Hi; •!•■•’>L 2.0V£S??2.01 .04O . ä q¥0 :öF \3T42 1 DJFi.q-z2:0OQ.O00.0;00 .0-D-fÄ210 -2 .0 . ____2.0 — 4 779 Ci' --------3T42 2 DJF StrainFigure A 19 a) Climatological distribution of precipitation (mm/d) for the DJF season basedon the analysis of Legates and Willmott (1990). b)4) Simulated distributions obtainedwith different versions of the ECHAM model. For model notations see Fig. A l . Contourspacings: .5, 1., 2., 4., 8., ... mm/d. Light (dark) stippling indicates areas with P<0.5(P>4) mm/d, respectively.1360.5.... 0 .-FB-o-s ra in '_2.0_> - - - - 201T21 DJFeo 5 il g -QO.02T21 DJF irain.02.0■'i-iv : \ $ i52.0- i 2 -°?rain3T42 1 JJAhöo ö'° ° - G -00 .00;5:8 ;6------ 2|0..U?.SFöi6-Figure A 20 As in Fig. Al 9 except for the JJA season.138yyiij.OOjO2.01.0-0 •A“”"'1- 3.7. .4...” ----------1 ______ -2:0 — 2 - 0- ~ i! .... -A zrs ox-2102.0 — T~2' ■5ram.2.0''2 .0 210-är .....••—-train3T21 JJAFigure A 20 (cont.)1391406s15.5532-361524t4-Figure A 21 Standard deviation of 850 hPa temperature (K) due to transient eddiescomputed from the unfiltered time series for the DJF season, a) ECMWF analyses, b)ECHAM1 , c) ECHAM2, d) ECHAM3/T21, e) ECHAM3/T42 (1 . decade), f) ECHAM3/T42 (2. decade).141bKe2.6202,2 33 ZFTF32 — —2.5.2— --2.5a-1dFigure 4 22 As in Fig. A21 except for the band-pass filtered frequency range.14232fefl?• E>bFigure A 23 As in Fig. A21 except for the low-pass filtered frequency range.143144a3-|—54-5-6236-4021Figure A 31 As in Fig. 30 except for the band-pass filtered frequency range.154"37asoIOCBO'4020.2080 6080•20 ■20-■BCFigure A 32 As in Fig. 30 except for the low-pass filtered frequency range.155156rva150350 / .61350j .cP66l' V150015 .2<510552143i30 ,25IP1510Figure A 37 As in Fig. A36 except for the band-pass filtered frequency range.162ba 20 0<0■30,/SO201020V?A«.211010 -----JTF“5 '? .io:312ii o21ro■2lcFigure A 38 As in Fig. A36 except for the low-pass filtered frequency range.1631645«o206,zo10 zi010165>3'20>56e5-.5.0.10Figure A 39 Northward transport of sensible heat (Km/s) by transient eddies at 850 hPacomputed from the unfiltered time series for the DJF season, a) ECMWF analyses, b)ECHAM1, c) ECHAM2, d) ECHAM3/T21, e) ECHAM3/T42 (1. decade), f)ECHAM3/T42 (2. decade).165Figure A 40 As in Fig, A39 except for the band-pass filtered frequency range.166a6'.I'S50,6'00 0..0e 0,fOi'O005Figure A 41 As in Fig. A39 except for the low-pass filtered frequency range.167168a2110 J ° Jo, 6■55 '&$06160'S10-10■0-1J>=.15I r/C\ ’’3Figure A 42 Northward transport of sensible heat (Km/s) by transient eddies at 300 hPacomputed from the unfiltered time series for the DJF season, a) ECMWF analyses, b)ECHAM1, c) ECHAM2, d) ECHAM3/T21, e) ECHAM3/T42 (1. decade) f)ECHAM3/T42 (2. decade).169crFigure A 43 As in Fig. A42 except for the band-pass filtered frequency range.1700-2*2 2,Figure A 44 As in Fig. A42 except for the low-pass filtered frequency range.171

1

Abstract

Extracted data

1

Abstract

Extracted data

Related items

Related items