1. Introduction

When the components of a coupled climate system model are joined together and freed from the constraints of fixed surface boundary conditions, the statistical state of the simulated climate diverges from the initial condition (and generally from the observed climate). This temporal development of systematic biases in the simulation is referred to as climate drift. The usefulness of climate system simulations in studies of natural climate variability or anthropogenically induced climate change is limited to some extent by such drift. Many of the climate signals we wish to detect and understand are relatively subtle, and drift in the simulation can mask or contaminate the sought-after signal. In addition to contaminating the analysis of transient climate signals, drift can lead a simulation into an entirely different climate regime than intended, for example, an ice-free earth, due to various feedback processes and the existence of multiple quasi-equilibrium states in the climate system.

Attention has been focused on two major sources of climate drift: lack of equilibrium in the initial condition of the component models (primarily the ocean) and incompatibilities in the fluxes across component interfaces. Previous studies (Moore and Gordon 1994) indicate that while initialization is an important factor, changes in the initialization procedure cannot overcome inherent shortcomings in the component models that lead to flux incompatibilities. That is, in current generation models, climate drift is an indication of deficiencies in the component model physics and thus an indication of what aspects of the system are in need of improvement. Pragmatic considerations have led some groups to apply flux correction techniques (Sausen et al. 1988; Meehl 1995) to mask some of the biases inherent in the component models in order to achieve stable simulations of the equilibrium climate close to the observed. This state of affairs has led some authors (Weaver and Hughes 1996) to conclude that “flux adjustments are inevitable using the present generation of AGCMS . . . .”

The National Center for Atmospheric Research (NCAR) Climate System Model (CSM) is a comprehensive model of the physical climate system. It comprises a general framework for coupling the system components called the “flux coupler” (Bryan et al. 1996), an atmospheric general circulation model (Kiehl et al. 1998), an ocean general circulation model (Gent et al. 1998), a dynamic–thermodynamic sea-ice model (Weatherly et al. 1998), and a land surface process model (Bonan 1998). (The details of each component and their systematic biases when run in uncoupled mode are described in the aforementioned references and other papers in this issue.) The purpose of this study is to describe the temporal development of the systematic errors (i.e., the climate drift) in a multicentury control integration of the CSM and, to the extent possible, identify the specific component model deficiencies that lead to them. In contrast to many other recent coupled model experiments, this integration has been performed without flux corrections.

For the purposes of this analysis we characterize drift according to the timescale of the response. First we will consider the fast timescale response (up to a few years) associated with adjustment of the upper ocean, sea ice, land surface, and atmosphere. In the present experiment these drifts are quite modest compared to many other recent coupled model experiments, both with (von Storch et al. 1997; Johns et al. 1997) and without (Coleman et al. 1995; Guilyardi and Madec 1997) flux corrections. Second, we will consider slower drifts on a timescale associated with adjustment of the properties of the ocean interior. In the present simulation there are significant drifts of this class that persist throughout the duration of the integration. As research interest turns toward interdecadal variability and the role of the ocean thermohaline circulation in the climate system, this class of drift becomes of foremost concern. Consequently, in this paper the main focus will be on this longer timescale drift.

2. Description of the model, initialization, and integration procedure

The details of each component of the coupled system are described in the references cited above. The particular configuration used in the experiment under consideration will briefly be described here along with the procedure used to initialize and integrate the experiment. Further details can be found in Boville and Gent (1998). The atmospheric component model is the NCAR Community Climate Model (CCM3) with T42 spectral truncation (approximately 2.8° latitude–longitude resolution) and 18 vertical levels. The land surface model (LSM) operates on the atmospheric model grid but includes subgrid-scale heterogeneity of surface types. The ocean model has 45 vertical levels (12.5 m thick at the surface) and horizontal resolution of 2.4° in longitude and between 1.2° and 2.3° in latitude (finest at the equator and in high latitudes, coarsest in midlatitudes). The dynamic–thermodynamic sea-ice model is based on the cavitating-fluid rheology of Flato and Hibler (1992) and is formulated on the same grid as the ocean model. The flux coupler accommodates heterogeneity of surface types at the atmospheric model grid scale. Surface fluxes that depend directly on the state variables of more than one system component are computed on the finer of the two grids by interpolating the state variables from the courser to the finer grid. The coupler then performs area-weighted averaging and merging of surface fluxes from the land, sea-ice, and open-ocean areas within each atmospheric grid cell in a conservative manner.

As is standard practice, in this experiment the component models are spun up separately in uncoupled mode prior to the start of the coupled experiment. However, a number of the details of the spinup procedure differ from those used by other groups (Boville and Gent 1998). A common method of spinning up the ocean model prior to coupling is to integrate the model for a long period [possibly using the acceleration technique of Bryan (1984)] with surface forcing specified as a Newtonian restoring type formulation in which the sea surface temperature (SST) and salinity are damped toward observed values. Even for an otherwise perfect coupled system, this technique will inevitably lead to climate drift when the ocean is coupled to the atmosphere. An ocean model forced in this manner cannotsimultaneously obtain the observed sea surface temperature and salinity distributions and the observed surface flux distributions. A perfect simulation of SST would result in vanishing surface heat flux, and a perfect simulation of surface heat flux would require the SST to deviate from the observed value by approximately 0.02° for each watt per square meter of surface flux. This problem can be overcome to some extent by relaxing the sea surface temperature toward an “effective” temperature (Haney 1971; Barnier et al. 1995) rather than the observed sea surface temperature. In our experiments we have taken another approach. We force the uncoupled ocean model using a formulation of the surface fluxes that is as close as possible to that used in the fully coupled system but using atmospheric state data and radiative fluxes provided by reanalysis and recent climatologies. The details of this formulation are given in Large et al. (1997) and Gent et al. (1998). As a result, the equilibrium solution of the uncoupled ocean model provides a realistic representation of both sea surface temperature and salinity, and surface fluxes (Gent et al. 1998; Large et al. 1997).

The spinup of the atmospheric model is standard with sea surface temperature and sea-ice distributions specified from the climatology of Shea et al. (1990). The ice model spinup is somewhat iterative. The sea-ice initial condition for these experiments is taken from an intermediate state of a previous run of the fully coupled system. In the present experiment, the land model is reinitialized with a prescribed state at the start of the coupled run.

The second unique feature of the initialization of this experiment is a “phased-in” coupling of system components. Prior to coupling to the atmosphere model, the ocean and sea-ice models are coupled and integrated for 60 years using a repeating 5-yr period archive of daily data for the atmospheric surface state variables, radiative fluxes, and precipitation from the uncoupled atmospheric model spinup. In this phase of the experiment, daily mean atmospheric data is presented to the coupler and surface fluxes are updated using the current model-predicted surface ice and ocean states. Some recent studies (Power 1995) have suggested that the absence of synoptic variability in the forcing of the ocean during spinup can lead to drift in the coupled integration. This phase of the spinup is designed, in part, to minimize this source of drift but, more importantly, to allow the ocean to adjust to the climatological state of the atmospheric model. During the first 50 years of this phase the ocean is integrated using the acceleration technique of Bryan (1984) so that the deep ocean is integrated for the equivalent of 500 years. To stabilize the accelerated system, a 14-day running mean filter is applied to the fluxes before they are used to force the ocean model. Note, however, that the fluxes are computed using the daily state data so that the full nonlinearity of the flux computation is retained. Also, during the first 25 years of this phase, absorbed rather than downward solar radiation from the atmospheric model spinup run is specified and the evolving surface albedo is ignored in computing the net surface flux. This has the effect of eliminating ice-albedo feedback during the early phase of the experiment. Experience with prior runs of the CSM suggested that this was a useful technique to help bring a poorly initialized ice model into the system without excessive loss of Southern Hemisphere sea ice. The sensitivity of the model to this procedure is described in more detail in section 5. During the final 10 years of this spinup phase the ocean model is run in synchronous mode (no acceleration) and the daily surface forcing is applied unfiltered.

The atmosphere and land surface models are then brought into the system to begin the fully coupled experiment. The atmosphere, land, and sea-ice models communicate with the coupler, computing updated surface fluxes, at relatively high frequency, usually once per hour. The fluxes computed by the coupler are averaged over a 1-day period before being passed to the ocean, which in turn provides updated SST to the coupler once per day. The full coupled system has been integrated for 300 years in this mode.

3. Drift in the surface climate

The temporal evolution of annual zonal mean sea surface temperature from the initial uncoupled ocean model solution through the 60-yr ice–ocean spinup phase and the first 60 years of the coupled run is shown in Fig. 1. In general, the surface temperature distribution is very stable through all phases of the experiment and the mean meridional gradient of surface temperature retains realistic values. There is a slight (less than 0.5°C) warming on the equator during the ice–ocean spinup, followed by a larger cooling in the coupled experiment. The equatorial response is discussed further below and in Kiehl (1998). In the transition from uncoupled ocean equilibrium to ice–ocean spinup, there is a warming of approximately 1.5°C between 40° and 45°N. There is an initial cooling of approximately 1°C near 55°N that diminishes somewhat over time. Most of the adjustment is complete within a few years of the transition. There is no discernible change in the surface temperature at the transition from accelerated to nonaccelerated phase of the ice–ocean spinup. In the transition between the ice–ocean spinup and the fully coupled simulation, there is cooling over most of the Northern Hemisphere with the largest changes centered on 20° and 45°N (which reverses the warming trend of the spinup phase). The high northern latitudes cool at both transitions so that the largest differences from the initial condition occur near 65°N. There is a slight warming (less than 1°C) south of 35°S. There is no further secular drift in the sea surface temperatures beyond the first few years of the experiment. These results can be contrasted with other recent coupled model experiments without flux correction such as that described in Guilyardi and Madec (1997), which undergoes strong warming of tropical SST, resulting in a permanent El Niño–like state, or Gregory and Mitchell (1997), in which the global mean sea surface temperature drops by more than 2°C during the first 300 years of integration.

The regional distributions of the temperature changes across the two experiment transitions are illustrated in Fig. 2. The midlatitude warming in the Northern Hemisphere between the uncoupled ocean equilibrium and the ice–ocean spinup phase is seen to occur most strongly in the Pacific Ocean along the Kuroshio extension. The same region undergoes a cooling on transition to the fully coupled experiment, resulting in a relatively small net SST bias along 45°N. There is a smaller positive adjustment in the Atlantic Ocean in the region of the Labrador Current between the uncoupled ocean equilibrium and the ice–ocean spinup experiment. Over the course of the ice–ocean spinup the cooling to the south and east of Greenland diminishes. During the coupled experiment there is strong interdecadal variability in the surface temperature in this region as described in Capotondi and Holland (1998, manuscript submitted to J. Climate).

The east–west gradient of SST across the equatorial Pacific Ocean is enhanced during the ice–ocean spinup with a warming of 0.5°–1°C in the warm pool region and cooling in the central and eastern equatorial Pacific. Again, this tendency is largely reversed at the transition to the fully coupled experiment with a strong cooling over the warm pool region. The resulting east–west gradient in SST along the equator in the coupled integration is realistic, but there is a basinwide cold bias of approximately 1°C (Meehl and Arblaster 1998). As described in more detail in Kiehl (1998), the warming during the ice–ocean spinup phase is in response to an overestimate of surface shortwave radiation. During the coupled run, the atmospheric model responds with an increase in zonal wind stress, increasing the latent heat flux and cooling the surface temperatures.

At the transition between the uncoupled ocean and the ice–ocean spinup experiment, slight warming is apparent in the eastern boundary upwelling regions off the coasts of North and South America and southern Africa. In each case the area of warming is greatly enhanced and the response amplified on transition to the fully coupled experiment. This warm bias is associated with a deficit in the simulated marine stratus clouds and excess solar radiation in these regions and has been identified in many other coupled model experiments (Meehl 1995). The warming during the ice–ocean spinup phase is moderated by the fixed atmospheric temperatures that contain some information on the true climatological sea surface temperature distribution. The effective infinite heat capacity of the atmosphere during this phase can compensate for the excess solar radiation by absorbing increased latent and sensible heat fluxes. This compensating mechanism is removed when the atmospheric model is coupled, resulting in a widespread warming over the eastern North and South Pacific and South Atlantic Oceans. Also, in the case of the regions off the coasts of Peru and California, the upwelling favorable alongshore wind stress decreases in strength, amplifying the warming.

There is relatively little change in temperature near Antarctica and correspondingly little change in the area of sea-ice coverage. This result is also in sharp contrast to many other coupled model solutions in which Southern Hemisphere sea-ice coverage drops well below observed levels (e.g., von Storch et al. 1997). During both transitions, there is cooling in the Barents Sea region and in the western Bering Sea. As will be described further below, while the areal extent and volume of sea ice remain realistic, there are problems associated with ice–ocean interactions in this experiment that lead to drift in the deep ocean.

Zonally averaged sea surface salinity as a function of latitude and time is shown in Fig. 3. In contrast to the situation with surface temperature, there are persistent drifts in zonal mean salinity and little tendency for compensation between the ice–ocean spinup and the coupled experiments. The strongest changes are seen to occur in the Arctic region where the surface salinity increases by several parts per thousand by year 60 of the coupled run. These changes are described more fully below. The surface salinity adjacent to the Antarctic continent also increases early in the ice–ocean spinup phase but stabilizes and shows little subsequent increase. The remaining surface ocean, equatorward of approximately 70° latitude in each hemisphere, undergoes freshening. The strongest decrease in surface salinity occurs between 10° and 20°S. This results from the split structure of the ITCZ that develops with excessive precipitation south of the equator in austral summer (Meehl and Arblaster 1998). The decrease in salinity at subtropical latitudes is in spite of increased net evaporation due to relatively strong trade winds in the CCM3. The decrease in surface salinity reflects a general depletion of salt from the upper ocean as described further below.

Other than an initialization transient in land surface temperature during the first decade of the coupled integration, there is essentially no drift in other aspects of the surface or atmospheric climate through the 300 years of the integration.

4. Drift in the ocean interior

The volume mean ocean potential temperature during the ice–ocean spinup and coupled integrations is shown in Fig. 4. The ocean is cooling throughout the coupled experiment at an average rate of 0.07°C per century, corresponding to an area-averaged surface heat flux out of the ocean of 0.35 W m⁻². The volume mean potential temperature drops more rapidly during the accelerated phase of the ice–ocean spinup. However, the area-averaged surface heat flux H, inferred from the change in heat storage corrected for the acceleration factor γ(z),

View Expanded

is approximately −0.5 W m⁻² after the first 10 years, not much larger than in the coupled experiment. We note that the magnitude of the flux imbalance is smaller than the signal expected from greenhouse warming scenarios, such as a doubling of CO₂.

The current version of the CSM lacks a river-routing scheme to return net continental precipitation to the ocean. Instead, an instantaneous balance is explicitly imposed between precipitation and evaporation over the ocean and sea ice by increasing precipitation by a globally uniform multiplicative factor (Boville and Gent 1998). Thus, in the absence of drift in the volume of sea ice there is no drift in the global ocean averaged salinity. However, there is redistribution of salt (and heat) within the ocean during the course of the integration. The evolution of potential temperature and salinity at two levels are shown in Fig. 5. In the main thermocline, salinity decreases at approximate 0.3 ppt per century during the ice–ocean spinup and first 50 years of the coupled integration, and at approximately 0.2 ppt per century subsequently, while temperature comes to equilibrium after approximately 70 years of coupled integration. In the deep ocean salinity increases at a nearly constant rate through both the spinup (by stretching the time axis by the acceleration factor the slopes can be directly compared in the accelerated and nonaccelerated periods) and coupled phases of the experiment. The potential temperature is nearly equilibrated at this depth by the end of the ice–ocean spinup phase of the experiment. However, this is due, in part, to a lower bound on temperature of newly formed deep water set by the freezing point of seawater. Examination of the full water column shows that cooling continues at shallower depths, with equilibrium reached at later times moving up from the bottom.

The source of the deep drift is clearly revealed in maps of temperature and salinity change during the spinup phase of the experiment. Within the first decade of the spinup experiment (Figs. 6a,b) increasing salinity is apparent in the western Ross Sea and Weddell Sea, accompanied by colder temperatures surrounding the Antarctic continent. By the end of the accelerated phase of the spinup experiment the entire circumpolar region has become saltier and colder. By the end of the ice–ocean spinup (Figs. 6c,d) it is also apparent that the deep water formed in the North Atlantic has increased in salinity as well as temperature, with the signal carried into the South Atlantic by the Deep Western Boundary Current. Thus, both of the primary water masses ventilating the deep ocean have higher salinity than in the uncoupled ocean equilibrium or observations. Subsequently, the colder, higher salinity deep water masses spread into the rest of the world ocean. At the end of the coupled experiment, the ocean below 4000 m is near the freezing point, approximately 2.5°C colder than observations (Fig. 7a) and the deep salinity has increased by 0.3 ppt (Fig. 7b). The vertical distribution of salinity is generally reversed: the subsurface (100 m) maximum is replace by a global mean surface halocline.

The decrease in temperature (Fig. 7c) and increase in salinity (Fig. 7d) of Antarctic Bottom Water (AABW) cause an increase in the density and production rate of this water mass. The streamfunctions for the annual mean meridional overturning circulation (Eulerian plus eddy induced) for the uncoupled ocean equilibrium, the final decade of the ice–ocean spinup, and the last 50 years of the coupled experiment are shown in Fig. 8. The sinking of water along the antarctic continent increases from less than 5 × 10⁶ m³ s⁻¹ in the uncoupled ocean equilibrium to more than 20 × 10⁶ m³ s⁻¹ during the ice–ocean spinup phase, diminishing somewhat by the end of the coupled experiment (but remaining much higher than the uncoupled ocean model equilibrium value). The increase in sinking is accompanied by increased transport of AABW across the circumpolar channel into the basins to the north (primarily the Indian and Pacific Ocean basins). A secondary effect of this is an acceleration of the Antarctic Circumpolar Current (ACC) through the interaction of these deep flows with topography. The ACC transport in the uncoupled ocean equilibrium is approximately 120 × 10⁶ m³ s⁻¹, close to the observed value. Within the first few decades (centuries of deep water integration) of the ice–ocean spinup, the ACC transport increases to 280 × 10⁶ m³ s⁻¹ and remains at more than twice its initial value throughout the coupled integration. Eastward momentum input by the wind stress is largely balanced by westward topographic form drag in the circumpolar channel (Bryan 1997; Gille 1997). The westward-directed form drag is proportional to the strength of the southward flow below the depth of the shallowest sills in the channel. The enhanced northward flux of AABW thus diminishes the strength of the topographic form drag, allowing the ACC to strengthen until bottom friction and lateral stresses can reestablish a balance. This process is discussed further in the following section.

From Fig. 8 we see that the sinking in the Northern Hemisphere (almost exclusively in the Atlantic) increases from the uncoupled ocean equilibrium value, which is close to observational estimates (Gent et al. 1998) as well. This can be understood in terms of the incompatibility of the meridional ocean heat transport in the uncoupled ocean model and that implied by the uncoupled atmospheric model (Fig. 9). While the heat transports are in good agreement in the Southern Hemisphere, the poleward heat transport implied by the uncoupled atmospheric model exceeds that of the uncoupled ocean model by up to 10¹⁵ W in the Tropics and northern subtropics. The increased meridional overturning during the ice–ocean spinup and coupled integrations serves to increase the ocean heat transport to a level closer to that implied by the uncoupled atmosphere model. The heat transport in the ice–ocean spinup and coupled integrations is further enhanced by an increased contribution of the horizontal gyre motions due to enhanced transport of the subtropical gyres (Danabasoglu 1998).

The strong drift in surface salinity in the Arctic was noted above. This drift extends into the ocean interior. In particular, the Arctic halocline is eroded away during the ice–ocean spinup and early part of the coupled integration. In this case we can identify at least two factors in the basic system design that could contribute to this problem. The two major sources of freshwater to the surface layer of the Arctic are inflow of low-salinity water through Bering Strait and river runoff (Aagaard and Carmack 1989). Both of these sources are missing from this experiment: Bering Strait is closed for reasons of computational economy and the coupled model lacks a river-routing system (Boville and Gent 1998). The uncoupled ocean equilibrium has a somewhat weak representation of the halocline that is maintained by the restoring of salinity values to the World Ocean Atlas 1994 climatology (Levitus et al. 1994; Levitus and Boyer 1994, hereafter WOA94) under regions diagnosed with sea ice. The restoring provides an ad hoc representation of these missing freshwater sources. In both the ice–ocean spinup and coupled integrations this restoring is removed, allowing the halocline to erode. Further, during the ice–ocean spinup phase the volume of Arctic sea ice increases yielding a net increase in salinity and during the coupled integration the growth of new sea ice within the Arctic and export through Fram Strait exceed observational estimates by about 50% (Weatherly et al. 1998). Thus, enhancement of freshwater sinks for the Arctic region compound deficient sources, leading to persistent salinity drift. Within 60 years of coupled integration the salinity in the Arctic is essentially uniform from surface to bottom.

5. Discussion

The difference between the uncoupled ocean and atmosphere model heat transport curves shown in Fig. 9 is a measure of the zonal annual average mismatch in surface heat fluxes, δQ,

View Expanded

where L_x is the ocean surface zonal width at latitude y, and T_A and T_O are the uncoupled atmosphere and ocean model meridional ocean heat transports, respectively. Also, δQ is equal to the zonal annual average of the heat flux correction that would have been applied in this system had we chosen to do so, and it is shown in Fig. 10. There are several things to note about this result in comparison to flux corrections actually applied in other recent coupled experiments. First, the magnitudes are rather modest, with an area-weighted rms of 10.5 W m⁻² and nowhere in excess of 25 W m⁻². This can be compared to the result presented in Weaver and Hughes (1996, their Fig. 12a), where values approach 100 W m⁻² in the circumpolar region. Second, there are roughly equal areas of limited meridional extent of positive and negative values in each hemisphere. Again, by comparison, the result of Weaver and Hughes (1996) has positive values over most of the Southern Hemisphere and everywhere south of 30°S. In consequence, the heat flux imbalances in the present system can be accommodated fairly locally without the need for large changes in the ocean circulation patterns.

The primary source of long timescale drift in the present simulation has been shown to be the formation of Antarctic Bottom Water at unrealistic rates and with unrealistic properties. This drift begins during the ice–ocean spinup phase of the experiment, pointing toward interactions of the sea-ice and ocean models as the source of this problem. Weatherly et al. (1997) show that the offshore transport of sea ice at 70°S in the coupled experiment exceeds an estimate based on satellite observations by a factor of 4. Early in the ice–ocean spinup the overestimate is even larger: nearly a factor of 8. The excessive transport is, in turn, attributed to higher than observed ice velocities and a larger than observed northward turning angle of the ice relative to the surface winds, rather than biases in ice thickness or concentration. This large offshore ice mass transport is supported by excessive new ice production and accompanied by large amounts of brine rejection near the Antarctic continent. The freshwater loss from the surface due to ice formation reaches 3 m yr⁻¹ during the coupled integration. Toggweiler and Samuels (1995) review a number of estimates of the freshwater budget for the Antarctic region. They suggest that ice formation must be less than about 0.5 m yr⁻¹. In the uncoupled ocean equilibrium, the restoring term for salinity is indeed quite close to this value. Thus, the large changes near Antarctica are the result of the disturbance to the ocean freshwater balance associated with the introduction of unreasonably large meridional transports by the ice model.

Several additional short experiments (Table 1) have been carried out to isolate this problem. First, an experiment (c001.01) using the standard CCM3 atmospheric forcing data but with no ice model was run. In this case, temperature and salinity are restored to the WOA94 climatological values in regions diagnosed as sea ice covered in the Shea et al. (1990) climatology, the same technique used in the uncoupled ocean equilibrium experiment. In this experiment there is virtually no change in the sinking rate along the Antarctic shelf (Fig. 11) or in the transport of the ACC (Fig. 12). This clearly excludes increased atmosphere–ocean stress in the CCM3 forcing data relative to the National Centers for Environmental Prediction (NCEP) reanalysis forcing data as the direct cause of the acceleration of the ACC.

A second possibility is that the increased atmosphere–ice stress in the CCM3 forcing causes excessive offshore transport of sea ice and correspondingly excessive brine rejection near the coast. This possibility is explored in an experiment (g006.38) in which the atmospheric state variables passed to the flux coupler are taken from NCEP reanalysis rather than from the CCM3 spinup (the radiative forcing from CCM3 is retained). In this case there is diminution of the sinking rate and ACC transport relative to the standard case by about 25%. This suggests that biases in the CCM3 surface wind distribution are a contributory, but not exclusive, cause of the drift.

Finally, after the completion of the coupled experiment, it was discovered that an unrealistically large value of roughness length for sea ice was used in calculating the atmosphere–ice stress in both the standard ocean–ice spinup and the coupled run (Boville and Gent 1998; Weatherly et al. 1998). A short experiment reducing this parameter to more realistic values (g006.37) showed that the intensification of sinking rate and ACC transport are delayed by a few years and are again diminished in amplitude. Weatherly et al. (1998) indicate that the sensitivity of the ice transport to the air–ice drag coefficient value may itself be overestimated in the present model due to the use of a linear rather than quadratic ice–ocean drag law. The linear ice–ocean drag does not increase rapidly enough at high ice velocities. There is no indication that the cavitating-fluid rheology itself is a contributing factor in this problem.

As described above, a novel aspect of our spinup procedure is the suppression of ice-albedo feedback during the first 25 years of the ice–ocean spinup phase of the experiment. This procedure was adopted in prototype versions of the CSM to avoid excessive melting of sea ice in the Southern Hemisphere during the first few years of the integration. The impact of this procedure is illustrated in Fig. 13. Its utility in the present experiment is equivocal. The sea-ice coverage is reduced in both hemispheres when ice-albedo feedback is activated at the initial time, with near-complete disappearance of ice in the Southern Hemisphere during summer. However, the standard procedure results in an overestimate of sea-ice extent compared to observations throughout the year. The two solutions appear equally biased in their simulations of the observed seasonal cycle in sea- ice extent. On the other hand, the initial condition for the ice model used here had already been allowed to adjust to the ocean model and CCM3 surface forcing in a previous coupled integration. Further, as described in Boville and Gent (1998), the fixed atmospheric surface temperatures from the CCM3 reflect an overestimate of the climatological ice distribution and serve to maintain sea surface temperatures at the freezing point in regions where even small concentrations of sea ice are observed to occur. Thus, while the results for this particular case may not provide a compelling argument for this procedure, it does appear to be a useful technique when initially coupling ocean and ice models. However, these results do suggest that a shorter period than 25 years would be sufficient to accomplish the desired result.

6. Conclusions

The results of this multicentury experiment demonstrate significant progress toward achieving realistic simulations of the climate system using coupled models without flux correction. The surface temperature distribution remains realistic through 300 years of integration with essentially no drift in global mean surface temperature. Other aspects of the atmospheric circulation and surface climate, such as sea-ice extent, remain realistic. The minimal drift in the surface climate can be attributed to a close agreement in the meridional heat transports of the uncoupled ocean model and that implied by the uncoupled atmospheric model.

However, the experiment described here is not without drift. The deep ocean cools and becomes saltier, while the upper ocean becomes fresher at relatively constant rates through the length of the experiment. We have demonstrated that the major component of the drift is due to vigorous ventilation of the deep ocean by unrealistically cold and salty Antarctic Bottom Water, which in turn results from unrealistically large freshwater transport by the sea-ice model. It is likely, though not demonstrated, that eventually the drift in the deep ocean would manifest itself at the surface, perhaps through a collapse of the Northern Hemisphere thermohaline circulation as a result of the depletion of salinity in the surface ocean.

We expect that improvements in the next version of the CSM sea-ice model will largely resolve this problem. The flexible nature of the CSM system makes it relatively easy to explore the ice–ocean interaction problem and test improvements in the sea-ice model without incurring the cost of running an active atmosphere model. Similarly, other aspects of the drift and systematic biases, such as the eastern boundary marine stratus cloud deficit problem, can be explored with more economical upper-ocean models.

We thus take issue with the conclusion of Weaver and Hughes (1996) that “flux adjustments are inevitable using the present generation of AGCMS . . . .” From these results it appears that ocean–atmosphere flux adjustments are evitable using currently available state-of-the-art models. Clear pathways toward improvements in the other component models suggest that flux adjustments may be avoided altogether, but this remains to be demonstrated.