The low-resolution CCSM2 revisited: new adjustments and a present-day control run

Abstract. The low-resolution (T31) version of the Community Climate System Model CCSM2.0.1 is revisited and adjusted by deepening the Greenland-Scotland ridge, changing oceanic mixing parameters, and applying a regional freshwater flux adjustment at high northern latitudes. The main purpose of these adjustments is to maintain a robust Atlantic meridional overturning circulation which collapses in the original model release. The paper describes the present-day control run of the adjusted model (referred to as "CCSM2/T31x3a") which is brought into climatic equilibrium by applying a deep-ocean acceleration technique. The accelerated integration is extended by a 100-year synchronous phase. The simulated meridional overturning circulation has a maximum of 14×10 6 m 3 s −1 in the North Atlantic. The CCSM2/T31x3a control run is evaluated against observations and simulations with other climate models. Most shortcomings found in the CCSM2/T31x3a control run are identified as "typical problems" in global climate modelling. Finally, examples (simulation of North Atlantic hydrography, West African monsoon) are shown in which CCSM2/T31x3a has a better simulation skill than the latest low-resolution Community Climate System Model release, CCSM3/T31.


Introduction
Paleoclimatic model experiments usually require long integration times either to reach climatic equilibria which differ from the present-day situation or to simulate long-term (e.g., millennial) climate trends and changes. In order to facilitate long integration times, low-resolution configurations of the fully-coupled NCAR (National Center of At-20 mospheric Research) Community Climate System Model CCSM have been released both for version 2.0.1 ("CCSM2/T31") and for version 3.0 ("CCSM3/T31"). In these socalled 'paleo versions', the horizontal resolution of the atmospheric component is given by T31 spectral truncation (3.75 • by 3.75 • transform grid), whereas the ocean has a nominal resolution of 3.6 • by 1.6 • with 25 levels.

EGU
In the framework of CCSM, atmosphere and land models share an identical horizontal grid. Likewise, ocean and sea ice use one and the same horizontal grid. In CCSM2/T31x3a, the ocean/sea-ice component is formulated on an orthogonal grid which shifts the north pole singularity into Greenland to avoid time-step constraints due to grid convergence. This grid is referred to as "gx3v4" (Fig. 3). It has a longi-5 tudinal resolution of 3.6 • . The latitudinal resolution of "gx3v4" is variable, with finer resolution (approximately 0.9 • ) near the equator.
In CCSM2/T31x3a, several adjustments to the standard CCSM2.0.1 release are applied. The overall goal is to amplify the Atlantic meridional overturning circulation. Due to the computational expense of performing fully-coupled experiments systematic sen- 10 sitivity studies, elucidating the effects of each modification separately, are not feasible for the time being. The tuning of CCSM2.0.1 is based on experience with other models. CCSM2/T31x3a includes the following adjustments: -The Greenland-Scotland ridge is slightly deepened such that the sill depths are ∼590 m and ∼900 m in the Denmark Strait and in the Iceland-Scotland passage, 15 respectively. Given the vertical resolution of the "gx3v4" ocean grid, these depths are in line with the real bathymetry. A deeper Greenland-Scotland ridge facilitates the exchange of water masses between the North Atlantic and the Nordic Seas where deep winter convection takes place (cf. Koesters et al., 2004).
-Background vertical mixing in the ocean is set to a constant value of 0.3 cm 2 /s. In 20 the default model set-up vertical background mixing increases from 0.1 cm 2 /s at the surface to 1.0 cm 2 /s at 5000 m depth (a value of 0.3 cm 2 /s is reached at about 2300 m depth). Thus, compared to the default setting, vertical mixing is slightly increased in the upper ocean below the surface boundary layer (alternatively, one could have applied geographically varying upper-ocean parameters with lower 25 vertical diffusivity in the tropics and much higher values in the Southern Ocean where internal wave activity is known to be enhanced, see Gnanadesikan et al., 2006). Vertical mixing provides a mechanism for the conversion of cold deep EGU waters into warm water of the upper layers. The crucial role of vertical mixing in driving the thermohaline circulation has been demonstrated in numerous studies (e.g., Bryan, 1987;Wright and Stocker, 1992;Marotzke, 1997;Prange et al., 2003).
-For the Redi and bolus parts of the Gent-McWilliams parameterization diffusivities 5 are set to 1.2×10 7 cm 2 /s. This represents a 50% increase compared to the default. It is expected that higher horizontal mixing counteracts halocline formation in the northern North Atlantic, thereby favouring convective activity and NADW formation (cf. Schmittner and Weaver, 2001).
-The coefficient used in the quadratic ocean bottom drag formula is increased from 10 10 −3 to 10 −2 . The most important effect of this change is a substantial retardation of the flow through the shallow Bering Strait. This throughflow is associated with an import of relatively fresh water from the North Pacific to the Arctic Ocean and the Nordic Seas, where it is likely to affect convective activity. It has been shown in several model studies that a reduction of the Bering Strait throughflow strengthens 15 the meridional overturning circulation in the Atlantic Ocean (e.g., Hasumi, 2002;Wadley and Bigg, 2002;Prange, 2003).
-At each ocean model time step, freshwater fluxes (precipitation plus river runoff) into the surface grid cells of the Arctic Mediterranean, Labrador Sea and Hudson Bay are reduced by 50%. The corresponding amount of freshwater is homo-20 geneously distributed over the entire Pacific Ocean (Fig. 4). This leads to an effective sea surface salinity increase in regions that are potentially important for NADW formation. Beyond these regions, the hydrological cycle is simulated without unphysical adjustments. This is a main advantage over the more common application of global flux-correction fields. The regional freshwater flux adjust-Introduction

EGU
In addition to the model tuning which aims at boosting the Atlantic overturning circulation, optimized sea-ice/snow albedos are applied based on results from stand-alone sea-ice model experiments: Maximum albedos for thick, dry sea ice are set to 0.82 and 0.38 for the visible and near-infrared spectral band, respectively. The near-infrared albedo for dry snow is set to 0.74. No distinction is made between the hemispheres.

Accelerated integration
Accelerated integration techniques are often applied to climate models to reduce the computational expense. In order to obtain a present-day climatic equilibrium, a deep-ocean acceleration technique -which is highly efficient in the framework of CCSM2/T31x3a -is employed here. This approach allows for increasing tracer time steps with depth, exploiting the relaxation of the Courant-Friedrichs-Lewy constraint due to diminishing current speeds in the deep ocean (Bryan and Lewis, 1979;Bryan, 1988). Such an asynchronous integration technique has proven useful for searching equilibrium solutions without any interest in the transient behaviour of the model: Once an equilibrium is reached (i.e., vanishing time-derivates in the model equations), the 15 solution is independent of the time-stepping. However, numerical acceleration techniques can severely distort the model physics. Two major concerns have been raised regarding asynchronous deep-ocean timestepping. Firstly, this approach does not ensure tracer conservation. Conservation of heat and salt is violated whenever vertical fluxes occur between neighbouring grid 20 boxes that solve the prognostic tracer equations with different time steps (Danabasoglu et al., 1996). Secondly, time-derivates never vanish in a realistically forced ocean model due to intra-and interannual variability. In order to quantify these errors, Danabasoglu (2004) recently applied accelerated integration methods to POP 1.4 subject to realistic forcing. Comparing equilibrium temperatures and salinities obtained 25 by deep-ocean acceleration with those from a 10 000-year synchronous control run, he found that errors are of order 0.1 K and 0.1 psu, respectively, provided that two condi-Introduction EGU tions are met: (i) vertical variations in time step are restricted to depths where vertical tracer fluxes (i.e. vertical gradients) are small enough that tracers are conserved well enough (in particular below the pycnocline); (ii) the accelerated integration is extended by a synchronous phase of -at least -several decades (Danabasoglu et al., 1996;Wang, 2001;Danabasoglu, 2004). The synchronous extension is important not only to correctly capture oceanic variability, but also to test the stability and reliability of the accelerated equilibrium solution (cf. Bryan, 1984; see also Sect. 4.1). Previous modelling studies have demonstrated the ability of acceleration techniques to reach an equilibrium paleoclimatic solution (see, e.g., Huber and Sloan, 2001;Huber and Nof, 2006).
The acceleration-induced small errors found by Danabasoglu (2004) are tolerable for most paleoclimatic applications. In particular, errors in large-scale oceanic mass and heat transports turned out to be negligible (for instance, the error in maximum Atlantic northward heat transport was about 0.01 PW or 1-2%). Using the same oceanic model grid as in the study by Danabasoglu (2004), a similar deep-ocean acceleration scheme 15 is used here. The surface time step in the ocean model is set to 1 h (time-step restriction due to numerical instability) and does not change down to a depth of 1300 m. Below 2500 m, the tracer time step is increased by a factor 50. Between 1300 m and 2500 m, the tracer time step has a linear variation.

Experimental design and spin-up
For the present-day control run of CCSM2/T31x3a, the atmospheric composition of 1990 AD is adopted. Volume mixing ratios of greenhouse gases are listed in Table 1

EGU
Using the deep-ocean acceleration technique described in the previous section, a climatic equilibrium can be achieved within a few centuries of integration. After a short (7 years) synchronous spin-up phase, depth-accelerated integration is applied for 293 years, followed by a centennial synchronous extension. This gives a total integration time of 400 surface years for the coupled climate model corresponding to 14 757 deep-5 ocean years. Only the third stage of the integration procedure (i.e. the centennial synchronous phase) shall serve for an evaluation of the simulated present-day climate. Figure 5 shows the time series of global ocean temperature over the accelerated spin-up phase and the synchronous extension. Initialized with climatological data (Steele et al., 2001) the ocean cools by about 0.6 K until reaching a near-equilibrium 10 state. During the same time, global average salinity increases by about 0.01 psu (not shown). This salinity drift is mainly attributable to the non-conservative character of deep-ocean acceleration.
The Hovmöller diagrams in Fig. 6 display the time evolution of zonally and meridionally averaged potential temperature and salinity for the Atlantic, Pacific, and Indian 15 oceans. It is clearly visible that the global oceanic cooling (Fig. 5) can be ascribed to a decrease in deep and bottom water temperatures, while the upper layers gradually warm during the spin-up. The abyssal potential temperature drift during the last century of accelerated integration (i.e. between surface year 200 and 300) is below 0.1 K, i.e. smaller than 2×10 −5 K per deep-water year. For the same time interval, abyssal 20 salinity changes are about 0.03 psu or 6×10 −6 psu per deep-water year. These rates are sufficiently small, confirming that the climatic state is at equilibrium for practical purposes at the end of the accelerated spin-up phase. Starting from an ocean at rest, most mass (or volume) transports obtain quasiequilibrium within 50 surface years. Figure 7 shows the temporal evolution of the 25 Atlantic meridional overturning streamfunction at 25 • S. At equilibrium, almost 12 Sv of deep water are exported to the Southern Ocean between 1000 and 3000 m depth; below 3000 m, 2-3 Sv of Antarctic Bottom Water (AABW) enter the Atlantic Ocean. The major goal of model tuning is achieved: CCSM2/T31x3a produces a robust overturning OSD 3,2006 The low-resolution CCSM2 EGU circulation in the Atlantic Ocean which induces a substantial northward heat transport (cf. Sect. 4.2.1). In this stable climatic mode, the northern high-latitude freshwater flux correction totals 0.107 Sv (averaged over the last 100 years of the integration period). The largest transport of water in the world ocean occurs within the Antarctic Circumpolar Current (ACC). The volume transport through Drake Passage rapidly equi-5 librates during model spin-up approaching 90 Sv (Fig. 8). It is important to note that the time series of oceanic volume transports provide a hint on the stability and reliability of the accelerated equilibrium solution (cf. Peltier and Solheim, 2004). Large-scale volume transports (like the meridional overturning circulation or the ACC) quickly respond to changes in the forcing, generally adjusting within a few decades (e.g., Gerdes 10 and Koeberle, 1995;Danabasoglu et al., 1996). If the accelerated integration led to a "false equilibrium", a rapid reorganisation of the oceanic volume transports would be expected after switching from accelerated to synchronous integration at year 300 (which is obviously not the case).

Ocean
For the following evaluation of the CCSM2/T31x3a present-day climatic equilibrium, the last 90 years of the synchronous integration phase are considered; that is, averages from surface years 311-400 are calculated and compared to observational climatologies or observation-based estimates.  et al., 2004), a too weak zonal wind forcing at ACC latitudes is most likely responsible for the shortcoming in CCSM2/T31x3a (cf. Sect. 4.2.3). The maximum meridional overturning strength in the North Atlantic appears to be in line with observation-based estimates. Note, however, that only 60% of deep-water formed in the North Atlantic is exported to the Southern Ocean ( Fig. 9). Accordingly, the Atlantic Ocean northward 5 heat transport simulated by CCSM2/T31x3a is at the lower end of the range suggested by observations. In addition, the flow of NADW is relatively shallow and the total formation of AABW is weak (Fig. 9). For comparison: Ganachaud and Wunsch (2000) estimate a northward AABW flow of 5-7 Sv into the Atlantic, 4-12 Sv into the Indian, and 5-9 Sv into the Pacific Ocean. 10 The horizontal distribution of ocean mean currents is displayed in Fig. 10. In the surface layer, the equatorial Pacific is dominated by Ekman-driven divergent flow. At 100 m depth, swift equatorial undercurrents, flowing eastward, are visible in all three oceans. In the Pacific Ocean, the Equatorial Undercurrent is supplied by meridional geostrophic inflow that compensates the Ekman transports, including inflows at the 15 western boundary. In the Indian Ocean, the eastward current is mainly fed by the South Equatorial Current which, in turn, is supplied by the subtropical gyre circulation and the ITF. In accordance with observations, the ITF receives water basically from the Pacific North Equatorial Current (cf. Gordon, 2001). The Atlantic Equatorial Undercurrent is mainly fed from the South Atlantic (South Equatorial Current). 20 The Benguela Current, appearing below the Ekman layer, separates from the African coast far too south. Similar problems arise with other eastern boundary currents (e.g., the Humboldt Current). Given the rather coarse resolution of the model grid, subtropical western boundary currents -including Kuroshio, Gulf Stream, East Australian Current, Mozambique Current, and Brazil Current -are simulated satisfactorily. As in reality, the 25 Brazil Current is conspicuously weak as compared with the other western boundary currents (cf. Peterson and Stramma, 1991).

EGU
At high southern latitudes, the flow field is dominated by the ACC. South of the ACC, the westward flowing Antarctic Coastal Current is simulated. Around the southern tip OSD 3,2006 The low-resolution CCSM2

EGU
of Africa, the model version of the Agulhas Current/leakage transports water from the Indian Ocean to the South Atlantic. This transport may be an integral part of the global conveyor belt circulation (Gordon, 1986). In the North Atlantic, a strong North Atlantic Current marks the boundary between the subtropical gyre and the cyclonic subpolar gyre. Providing the convective regions south of Greenland and in the Nordic Seas 5 with warm and salty water, the simulation of the North Atlantic Current is of utmost importance for the thermohaline circulation.
The flow field at 2000 m depth is characterized by a vigorous circulation around Antarctica. In the Atlantic Ocean, the southward movement of NADW constitutes the lower limb of the thermohaline overturning circulation. The NADW flow path forms an 10 anticyclonic loop in the North Atlantic, which has no counterpart in observations. South of 30 • N, the southward flow of NADW is confined to the Deep Western Boundary Current.
Potential temperatures simulated by CCSM2/T31x3a are shown in Fig. 11 along with observational data. Modelled sea surface temperatures (SST) in the tropical Indian 15 and Pacific oceans are lower than observed. This cold bias is up to 2 K in the equatorial central Pacific. The cold surface temperatures are associated with a larger than observed low-level cloud cover over the equatorial Pacific (not shown). In the tropical Atlantic, the western warm pool is too cold and the zonal SST gradient has the wrong sign. The tropical cold temperature bias is also visible at 100 m depth. The most 20 pronounced deficiencies at subtropical latitudes are found in the eastern boundary currents and major upwelling regions (along the west coasts of North America, South America, northwest Africa, and southwest Africa), where surface and subsurface temperatures are too warm. In northern high latitudes, the North Atlantic Current provides for moderate water temperatures south of Iceland and in the Norwegian Sea. Com- Figure 12 shows global salinity fields. The success of the CCSM2/T31x3a model adjustments is most evident when comparing the field of annual-mean sea surface salinity with that from the standard CCSM2.0.1 control run (Fig. 2). The entire North Atlantic, including subtropical and subpolar regions as well as the Nordic Seas and the Arctic Ocean, exhibits surface salinities which are now much closer to observations. 10 However, low-salinity water still caps off the Labrador Sea, forcing convection to occur further to the east. The reason for this shortcoming is unclear. Deficiencies in the wind stress curl, however, are likely to play a crucial role in the formation of the Labrador low-salinity cap (Gnanadesikan et al., 2006). In the Nordic Seas and northern North Atlantic, winter convection and, hence, deep-water formation takes place where upper-15 ocean salinities are around or above 35 psu.
In the South Atlantic, the model exhibits an upper-ocean fresh bias. The subtropical front is marked by the 34.9 psu isohaline at 100 m depth. In observations, the front resides well to the south of the Cape of Good Hope and the Australian continent. In CCSM2/T31x3a, the subtropical front is shifted far to the north (cf. Fig. 12). Part of 20 this fresh bias can be attributed to excessive rainfall between 35 • S and 60 • S (see Fig. 19). In the southeastern Atlantic, a possible source of error is the lack of Agulhas eddies that transport salty water into the Atlantic. A much finer grid resolution would be required to simulate the formation of these eddies. Relatively

EGU
saltier NADW in the model compared to observations. Traces of Eurafrican Mediterranean Water are absent at 2000 m depth in both the temperature (Fig. 11) and salinity ( Fig. 12) fields of the model.

Sea ice
Maximum and minimum sea-ice conditions in the northern and southern hemispheres 5 simulated by CCSM2/T31x3a are displayed in Fig. 13. The use of optimized seaice/snow albedos in CCSM2/T31x3a leads to an ice thickness of 2.5-3.5 m over the central Arctic Ocean. North of Greenland the sea-ice thickness increases up to 6 m. These results are in good agreement with upward-looking sonar observations (e.g., Bourke and Garrett, 1987) and satellite altimeter measurements (Laxon et al., 2003). 10 In the Arctic Ocean proper, the largest discrepancy between model and data is found along the East Siberian coast, where the model predicts ice thicknesses similar to those north of Greenland, while observations suggest thin ice (<1 m) or even ice-free conditions during specific summer months. The overly thick ice cover along the East Siberian coast can mainly be attributed to a deficient wind-stress forcing. A negative 15 sea-level pressure (SLP) bias over Alaska and northwestern Canada (cf. Fig. 16) forces sea ice to drift from the North American coast towards East Siberia, thus maintaining an unusually thick ice-cover in that region. In the North Atlantic and the Nordic Seas, the model produces too much ice area. Most severly affected in winter are the Labrador Sea, the regions east and northeast of Iceland, the sector south of Svalbard as well 20 as the western Barents Sea. The model reproduces year-round ice-free conditions over almost the entire Norwegian Sea. During the summer months, CCSM2/T31x3a simulates too much sea ice south of Greenland, around Svalbard, and in Baffin Bay.
In the Southern Ocean, the overall pattern of sea-ice cover is simulated satisfactorily. Although the model produces excessively thick ice along the eastern coast of the 25 Antarctic Peninsula, the typical "boomerang shape" thickness distribution in the Weddell Sea is qualitatively captured (cf. Strass and Fahrbach, 1998). As in the northern hemisphere, CCSM2/T31x3a exhibits a bias towards extensive sea-ice cover in the 1306 December-February (DJF) SLP simulated by the model against NCEP reanalysis data. The core positions of subpolar lows and subtropical highs are generally well captured in CCSM2/T31x3a, although the centers of the Icelandic Low and Azores High are slightly displaced eastward relative to observations. The strengths of the subtropical highs are overestimated in the northern hemisphere, and underestimated in the southern hemi-10 sphere. Anomalously high pressure is found in Arctic and sub-Arctic regions, where the model produces too much sea ice and too cold surface air temperatures (Labrador Sea, Greenland Sea, Barents Sea). Over Canada, the simulated winter pressure is lower than observed by up to 9 hPa. In high southern latitudes, the model exhibits a low pressure-bias over the subpolar seas, and a high-pressure bias over the Antarctic 15 continent. During June-August (JJA) the simulated strengths of subtropical highs are close to reanalysis data in the southern hemisphere (Fig. 15). In the northern hemisphere, CCSM2/T31x3a exhibits a pronounced high-pressure bias over the midlatitude oceans. In the Arctic realm, the simulated SLP is more than 6 hPa larger than 20 in NCEP data with a maximum deviation over Greenland. Antarctic and sub-Antarctic regions in the model climate are characterized by a strong low-pressure bias relative to reanalysis data. A similar seasonality of the Antarctic SLP bias (high pressure during DJF, low pressure during JJA) has been found in other climate models (e.g., Min et al., 2004). It should be noted, however, that errors in the reanalysis data cannot be 25 excluded for these extreme regions.
The deviation in simulated annual-mean SLP (Fig. 16) is associated with anomalously weak westerlies at the latitude of Drake Passage relative to NCEP data. EGU simulation of Southern Ocean wind forcing may partly be responsible for the low volume transport in the ACC. Moreover, it may partly account for the model's bias towards a weak Atlantic meridional overturning circulation (cf. McDermott, 1996;Gnanadesikan, 1999). The geographical pattern of DJF 2-m air temperature over land is shown in 5 Fig. 17. On global average, the simulated DJF temperature is 0.28 K greater than the observation-based  estimate. The winter surface climate of CCSM2/T31x3a is too warm over Greenland, northeastern Asia, and northern North America. The North American warm bias is associated with a low SLP anomaly (Fig. 14). During the summer season, simulated air temperatures are in better agree-10 ment with observations, and the global average is only 0.16 K warmer (Fig. 18). An overall cold bias for African and South American climates, however, is visible in both seasons. The same holds true for a pronounced warm bias over Antarctica. On annual average, the global 2-m air temperature over land is 0.1 K larger in CCSM2/T31x3a than in observations, while the global root-mean-square error (rmse) amounts to 15 3.35 K. A comparison of the simulated geographical distribution of annual-mean precipitation rate with CMAP observations is displayed in Fig. 19. The warm-biased region of northwestern North America receives excessive precipitation in CCSM2/T31x3a. The same holds for northeastern Siberia, albeit with a smaller magnitude of the error. Pronounced 20 wet biases are also visible over the central and southern parts of Africa, northern China, southern India, and eastern Indonesia, while Mainland Southeast Asia is too dry in the model. On the eastern side of the tropical Pacific, the model underestimates precipitation over central America and northern South America, while the coastal areas of Peru and Ecuador are too wet. Over the tropical ocean, the difference plot between model 25 and data reveals several shortcomings in the simulation: an east-west dipole over the Indian Ocean, a north-south dipole over the equatorial Atlantic owing to a rather diffuse Atlantic Intertropical Convergence Zone (ITCZ) in the model annual average, and a "double ITCZ" in the eastern Pacific. The "double ITCZ" emerges from a spurious Introduction EGU zonal band of excess rainfall just south of the equator, whereas the observations reveal a maximum extending from the west Pacific warm pool south-eastwards towards French Polynesia (the South Pacific Convergence Zone). Figure 20 shows the mean annual cycle of zonally averaged precipitation as derived from the model and CMAP data. While the major meridional shift in observed 5 tropical precipitation from the southern to the northern hemisphere takes place from March to April, it occurs between May and July in the model. During that time, zonalaverage CCSM2/T31x3a precipitation shows a false double structure of the ITCZ. In observations, the zonally averaged precipitation rate has a northern hemisphere maximum from June to August. In the model, the northern hemisphere maximum occurs 10 in September and is somewhat smaller than observed. In northern hemisphere midlatitudes, the seasonal variation of precipitation is overestimated by the model. In the southern hemisphere, CCSM2/T31x3a has a year-round dry bias around 30 • S, and a wet bias around 50 • S (see also Fig. 19).

Total heat transport 15
Annual averaged meridional heat transports by the ocean, the atmosphere, and the coupled system are displayed in Fig. 21 and compared with NCEP-derived values. In CCSM2/T31x3a, the maximum meridional ocean heat transport is 1.3 PW in the northern hemisphere, and 1.2 PW in the southern hemisphere. These transports are about 0.5 PW smaller than NCEP-derived values. Maximum meridional heat transports in the 20 atmosphere model are 4.9 PW and 5.4 PW in the northern and southern hemisphere, respectively. While the northern-hemisphere value is in good agreement with reanalysis, the southern hemisphere atmospheric transport is about 0.5 PW larger than the NCEP-derived value.
OSD 3,2006 The low-resolution CCSM2 Tropical climate variability on the short-range timescale from a few months to several years is dominated by the El Niño/Southern Oscillation (ENSO). Figure 22 shows the wavelet power spectrum of the Niño-3.4 index (SST 5 lated from the synchronous integration phase of the CCSM2/T31x3a control run. The global wavelet power spectrum exhibits a maximum around 2 years, while the ENSO period deduced from observational data has a broader spectral peak near 3-7 years. In harmony with observations, tropical interannual variability strongly varies from decade to decade (cf. Latif, 1998). It has been shown by Latif et al. (2001) and AchutaRao and Sperber (2002) that many climate models are not capable of simulating ENSO's phase locking to the annual cycle. To test the skill of CCSM2/T31x3a in simulating the seasonal cycle phase- 25 locking, the interannual standard deviations of the Niño-3.4 SST anomalies are calculated as a function of calendar month (Fig. 24). Although CCSM2/T31x3a simulates a secondary maximum in August, the strongest variability occurs during boreal win-1310 hemisphere winter climate. Here, the leading large-scale pattern associated with the NAO is extracted by principal component analysis on the winter 500-hPa geopotential height field, considering a limited spatial domain (90 • W-30 • E, 20 • N-80 • N). Figure 25 shows the leading empirical orthogonal function (EOF) obtained from the synchronous integration phase of the CCSM2/T31x3a control run. The first EOF accounts for 58.5% 10 of the total 500-hPa geopotential height variance over the spatial domain. This number is somewhat higher than the value calculated from NCEP reanalysis data (49.4%).
The 500-hPa height pattern consists of two centers-of-action. The northern center-ofaction is captured well by the model. The southern center is less well simulated, being displaced too far west and too far south over the Atlantic Ocean. 15

Discussion
CCSM2/T31x3a produces an overall reasonable present-day global climate. Nevertheless, the evaluation of the control run has revealed several shortcomings. Most of these shortcomings are well known as "typical problems" (i.e. common biases) in global, nonflux-corrected climate models. A strong surface cold bias in the equatorial Pacific, a 20 wrong sign of the tropical Atlantic zonal SST gradient, and positive SST biases at the eastern boundaries of the subtropical Pacific and Atlantic ocean basins (coastal upwelling regions of North/South America, northwestern/southwestern Africa) were to be expected from the history of ocean climate modelling (Mechoso et al., 1995;Latif et al., 2001;AchutaRao and Sperber, 2002;Davey et al., 2002;Wittenberg et al., 2006).

EGU
Sensitivity experiments suggest that errors in both surface solar radiation (through an under-prediction of stratus clouds in the atmosphere model) and wind stress ocean forcing (driving the coastal upwelling of cold thermocline water through surface Ekman divergence) each contribute about one-half to the common eastern boundary SST bias in climate models (Kiehl and Gent, 2004;Large and Danabasoglu, 2006). The repre-5 sentation of narrow coastal upwelling is also strongly dependent on the spatial resolution of the oceanic model grid. However, increasing the resolution of the ocean model does not necessarily reduce the SST biases in coastal upwelling regions (Yeager et al., 2006). It is important to note that eastern boundary surface biases are probably not confined locally. They can rather be advected over large distances and, hence, may 10 exert large-scale, remote influences over the coupled solution, even contributing to the double ITCZ problem (Li et al., 2004;Kiehl and Gent, 2004;Large and Danabasoglu, 2006).
The rainfall double ITCZ is a common problem in coupled non-flux-corrected climate models (Mechoso et al., 1995;Lambert and Boer, 2001;Harvey, 2003;Covey 15 et al., 2003;Li et al., 2004;Dai, 2006). A better simulation of the surface hydrography in the east Pacific coastal upwelling regions might improve the spatial structure of tropical rainfall. Recently, Zhang and Wang (2006) demonstrated that the use of a modified Zhang-McFarlane convection scheme significantly mitigates the double ITCZ problem in CCSM3.0, also resulting in an improvement of the Pacific SST simulation. 20 The annual-mean dry/wet bias over eastern/western equatorial Indian Ocean and a northern hemisphere mid-latitude wet bias in winter are other typical precipitation errors present in many coupled models (Lambert and Boer, 2001;Covey et al., 2003). Likewise, annual warm biases over Antarctica and Greenland as well as winter (DJF) warm biases over northeastern Asia and northern North America are often found in 25 climate models (Lambert and Boer, 2001;Covey et al., 2003).
The movement and distribution of sea ice is strongly determined by the high-latitude wind field. Excessive ice build-up along the Siberian coast is mainly attributable to an erroneous Arctic wind field and has been identified to be another common problem in SST variability in the equatorial Pacific too far to the west (Latif et al., 2001;Davey et al., 2002;AchutaRao and Sperber, 2002). Although the latest climate models tend to be more realistic in representing the frequency with which ENSO occurs, and they are better at locating enhanced SST variability over the eastern Pacific (van Oldenborgh et al., 2005;AchutaRao and Sperber, 2006), CCSM2/T31x3a's skill to simulate interannual 10 variability in the tropical Pacific is well within the range of other models. Teleconnection patterns associated with ENSO in CCSM2/T31x3a will be analysed elsewhere.
None of the above mentioned problems disappears in the higher-resolution (T42) version of CCSM2.0.1 or in the latest model release, CCSM3.0. Basically, control runs of CCSM2/T42, CCSM3/T31, CCSM3/T42 (and even CCSM3/T85) still suffer from the 15 same shortcomings with respect to precipitation, sea/land surface temperatures, seaice distribution (in both hemispheres) and tropical climate variability (see Kiehl and Gent, 2004;Yeager et al., 2006;Holland et al., 2006;Deser et al., 2006;DeWeaver and Bitz, 2006). Even though the winter surface warm bias over northeastern Asia and northern North America is reduced in CCSM3/T31 compared to CCSM2/T31x3a, the 20 errors are similar in the higher resolution versions CCSM3/T42 and CCSM3/T85 (see http://www.ccsm.ucar.edu/experiments). Table 3 summarizes errors of some globally averaged climatological quantities for different versions of the Community Climate System Model. Different climate variables are simulated with different levels of success by the different models and no one model is best for all variables. 25 While CCSM2/T42, CCSM3/T31, and CCSM3/T42 do not include any flux adjustments, the main disadvantage of CCSM2/T31x3a is the need for a freshwater flux adjustment at high northern latitudes in order to produce a robust Atlantic meridional overturning circulation. However, since this flux adjustment is limited to the Arctic Mediter- EGU ranean, Labrador Sea and Hudson Bay, the model behaves like a non-flux-corrected model with respect to tropical and subtropical climate dynamics and variability. This is particularly crucial when ENSO dynamics are considered (cf. AchutaRao and Sperber, 2002;Davey et al., 2002). It is important to note that the implementation of a freshwater flux adjustment is 5 not a step backwards compared to the older version of the climate system model, CSM1, which is often considered as a "non-flux-corrected" model (e.g., Latif et al., 2001;AchutaRao and Sperber, 2002;Covey et al., 2003;Stephenson and Pavan, 2003). However, CSM1 has no river runoff scheme. Instead, the precipitation over the ocean is multiplied by a factor to allow for a global surface freshwater balance. In so doing, CSM1 gets rid of the huge input of river water into the Arctic Mediterranean which -as the largest source of freshwater to the northern high-latitude seas -is about 0.1 Sv (Prange and Gerdes, 2006, and references therein). Therefore, at high northern latitudes, CSM1's simplification of the hydrological cycle has a very similar effect on the oceanic freshwater forcing than the flux adjustment applied to CCSM2/T31x3a.

Conclusions
The low-resolution version ("paleo version") of CCSM2.0.1 has been revisited. The original release has been adjusted by deepening the Greenland-Scotland ridge, changing oceanic mixing parameters, and applying a regional freshwater flux adjustment at high northern latitudes. The overall goal of these model adjustments, i.e. the improve-20 ment of the Atlantic meridional overturning circulation, has been achieved.
In the present paper, important aspects of the present-day control run of the adjusted version, CCSM2/T31x3a, have been critically analysed, and major shortcomings have been exposed. This provides a basis from which to judge numerical experiments performed with CCSM2/T31x3a. Most biases found in the CCSM2/T31x3a control run have been identified as "typical problems" in global climate modelling. Given its good simulation skills and its relatively low resource demands, CCSM2/T31x3a shows  10.1007/s00382-006-0144-6, 2006. Yoshimori, M., Stocker, T. F., Raible, C. C., andRenold, M.: Externally-forced