a. Coupled model

The overall structure of the NCAR Climate System Model and the control integration are described in detail in Boville and Gent (1998). The coupled model consists of individual component models for the atmosphere (Kiehl et al. 1998), ocean (Gent et al. 1998), land (Bonan 1998), and sea ice (Weatherly et al. 1998) all linked through a flux coupler (Bryan et al. 1996). The turbulent air–sea fluxes (e.g., sensible and latent heat flux) are calculated in the flux coupler given the surface-state variables (e.g., air and sea temperature) from the physical models. The coupler also partitions flux variables (e.g., solar radiation and precipitation) between the ocean, ice, and land surfaces. The present CSM version does not incorporate river runoff from the land, and the global freshwater budget is achieved by fractionally increasing the global precipitation rate, in effect redistributing river discharge to high rainfall regions such as the ITCZ. Although this approach does guarantee closure of the water budget, it clearly introduces some unphysical aspects to the solution. Fortunately for our purposes, the magnitude of the correction to precipitation is small (order 10%), and a full runoff model is being prepared for future CSM releases.

The ocean model component of the CSM is a coarse-resolution, global general circulation model specifically designed for studying climate processes on seasonal to decadal/centennial timescales (Gent et al. 1998). A variant of the z-coordinate, National Oceanic and Atmospheric Administration Geophysical Fluid Dynamics Laboratory Modular Ocean Model, the version of the model used in CSM has 45 levels in the vertical and a horizontal grid 1.2°–2.4° in latitude and 2.4° in longitude, with increased resolution at the equator and higher latitudes.

The NCAR ocean model contains a series of modifications (NCAR Oceanography Section 1996) relevant for surface forcing and water mass transformations: a quasi-adiabatic, isopycnal-oriented, mesoscale tracer transport parameterization (Gent and McWilliams 1990); the nonlocal, K-profile parameterization surface boundary layer mixing scheme (Large et al. 1994); and a bulk-flux forcing scheme for uncoupled solutions that more closely resembles the form of the ocean air–sea fluxes that occur during full atmosphere–ocean coupling (Large et al. 1997). An active sea ice model is included in the CO simulations (Weatherly et al. 1998), with thermodynamics based on the Semtner (1976) model and dynamics derived from a cavitating fluid ice rheology (Flato and Hibler 1992).

The CO climatology is constructed by averaging over a representative 10-yr period (years 90–99) of the simulation. The decade from years 90–99 is sufficiently far from the beginning of the integration to avoid the small initial coupling transient in surface properties. A comparison of the years 90–99 climatology with other 10-yr periods over the integration suggests that the years 90–99 surface properties are typical of CSM coupled solution and will be the main focus of our discussion. A notable exception is the subpolar North Atlantic region, which contains substantial variability on the 50–100-yr timescale in the CSM control integration (Capotondi and Holland 1997, manuscript submit to J. Climate).

The ocean component model stores monthly averages for surface properties and air–sea fluxes on the ocean model grid. The heat and freshwater fluxes on the ocean history files contain the effect of ice accretion onto the base of the sea ice and sea-ice melting but not the formation of sea ice due to subfreezing ocean surface layer temperatures, which is stored in the ice history files. The latter variable is also used for the ice melt potential when temperatures are above freezing (Bryan et al. 1996), and unfortunately the monthly averaging of ice model history variables during the control integration can lead to biased, low estimates of the ice formation rate, particularly near the sea-ice margin, where both melting and freezing occur in the same month.

Corrected ice formation rates are calculated by comparing zonal averages of the available monthly average ice formation rate with the zonal average oceanic freshwater residual computed from the incomplete net surface flux, meridional transport divergence, and ocean storage term (Fig. 1). If the ice formation rate were stored correctly, it would equal the oceanic residual, but the residuals can be up to a factor of 2–3 higher near the ice margins. The average ratio of the two terms computed monthly and by basin for the model decade 90–99 is then applied to local, monthly ice formation rates.

b. NCEP-STR climatology

A monthly surface flux climatology, NCEP-STR, is constructed using a bulk-flux scheme, a prescribed surface atmosphere state from the NCEP–NCAR reanalysis, and observed SST from the Shea et al. (1990) climatology. The required forcing datasets consist of near-surface wind speed, air temperature, and humidity, cloud cover, solar insolation, precipitation, and climatological SST and sea surface salinity (SSS) (NCAR Oceanography Section 1996). The Shea et al. (1990) and Levitus (1982) surface temperatures and salinities are used for the NCEP-STR water-mass transformation calculations.

The turbulent heat flux components (sensible and latent heat flux) and the net longwave flux are computed using standard bulk flux formula and climatological SST. The atmospheric variables are taken from a monthly mean atmospheric surface climatology (1985–88) compiled from the 6-h NCEP–NCAR reanalysis (Kalnay et al. 1996) and daily International Satellite Cloud Climatology Project (ISCCP) satellite cloud (Rossow and Schiffer 1991) datasets for 1985–88. The NCEP–NCAR reanalysis surface variables (wind speed, air temperature, and humidity) are affected by the assimilation of observational data but are also strongly influenced by the NCEP model, and as such they are classified as type B fields by Kalnay et al. (1996). The solar radiation and precipitation fluxes are specified from the daily ISCCP (Bishop and Rossow 1991) and monthly Microwave Sounding Unit (MSU) (Spencer 1993) satellite data products, respectively. The heat and freshwater fluxes are linked via the evaporation/latent heat flux term. Sea-ice distribution is diagnosed from the NCEP-STR SST data, and the heat and freshwater fluxes are set to zero under ice.

Considerable uncertainty is associated with current air–sea heat and freshwater flux estimates (Weller and Taylor 1993), and the atmospheric forcing data have been adjusted, within acceptable error ranges, to generate a globally balanced heat and freshwater budget for the ocean with the Shea et al. (1990) realistic sea surface temperature (NCAR Oceanography Section 1996; Large et al. 1997). The use of monthly atmospheric climatologies rather than the original 6-h data, which was chosen to match the manner in which the UN model is forced, introduces a relative error in the latent and sensible fluxes of less than about 10% (Esbensen and Reynolds 1981).

The uncorrected global heat balance results in net average warming of nearly 50 W m⁻². Two adjustments are applied—a reduction of the ISCCP surface insolation by 12.5% and a uniform decrease of the NCEP 2-m humidity by 7%—that contribute approximately equally to attaining global balance (Large et al. 1997). The humidity correction is based on an analysis of North Atlantic Voluntary Observing Ships observations that show systematic positive biases in dry–wet bulb estimates of the dewpoint temperature (Kent et al. 1993) and latent heat flux (Kent and Taylor 1995).

The solar correction is motivated by a number of observations (Cess et al. 1995; Pilewskie and Valero 1995; Ramanathan et al. 1995) that the cloudy atmosphere absorbs more solar energy than predicted by traditional atmospheric radiation transfer schemes such as used by Bishop and Rossow (1991) to compute the ISCCP surface data. The existence of this so-called excess absorption is a point of ongoing controversy (e.g., Bishop et al. 1997; Zender et al. 1997), and the possible global patterns over the ocean are not known. A uniform first-order adjustment is applied, therefore, independent of cloud cover (NCAR Oceanography Section 1996). The NCEP-STR precipitation data is increased by 13.6% to balance the global evaporation.

c. Uncoupled model

The UN surface properties and fluxes are taken from the monthly averages of the equilibrium state of the uncoupled ocean model spinup forced with the same bulk-flux forcing scheme as in the NCEP-STR climatology. A weak, regional restoring flux to the Levitus (1982) climatological surface salinity is also included in open ocean areas. Under-ice fluxes are specified with traditional strong restoring to observed temperature and salinity fields (Levitus 1982; Shea et al. 1990) where the sea-ice fraction is diagnosed from the Shea et al. (1990) SST climatology. A global freshwater balance is achieved by increasing the MSU precipitation by 8.7% to match the integrated evaporation rate. River inflow is not explicitly included but is implicit in the global precipitation factor and regional salinity restoring patterns. The local feedbacks for model sea surface salinity are weaker and are dominated by the model restoring term. Small deviations in the model SSS field arise but are generally damped back toward the climatological values.

The surface heat flux parameterization outlined by Haney (1971) and later refined by Han (1984) is a linearized version of our method. The apparent coupling coefficients between the net heat flux and sea surface temperature for such a method are large, 40–60 W m⁻² K⁻¹, and vary with wind speed and surface humidity (Oberhuber 1988; Barnier et al. 1995). With the atmospheric state specified (infinite heat capacity), the model sea surface temperature is, therefore, constrained to be similar to the observed temperature values both in the annual mean and seasonal amplitude (Large et al. 1997).

d. Water mass transformations

Air–sea heat and freshwater fluxes act to convert ocean surface waters from one density class to another, effectively driving a cross-isopycnal mass flux at the surface (Tziperman 1986). Following Speer and Tziperman (1992), the surface density flux f_ρ is given by

View Expanded

where c_p is the specific heat, and α and β are the thermal expansion and saline contraction coefficients. The water-mass transformation rate F(ρ) for the density interval ρ to ρ + Δρ is then found in discrete form by integrating f_ρ over the outcrop area between the two isopycnals:

View Expanded

where the δ function is zero except in the interval ρ to ρ + Δρ. The width of the density interval is set to σ_θ 0.1 kg m⁻³, and F(ρ) is expressed in Sverdrups where 1 Sv = 10⁶ m³ s⁻¹.

The annual mean water-mass transformations are calculated from the model and climatological datasets by ocean basin defined broadly as Atlantic–Arctic, Indian–Pacific, and Southern Ocean, taken as everything south of 35°S. The haline and thermal components to the water-mass transformation are also computed by setting α and β to zero, respectively, in Eq. (1). The water-mass transformation rates F(ρ) can be of either sign, depending on whether the surface fluxes are tending to increase (positive) or decrease (negative) surface density, and they give the direction and flow of water through a density class; the net water-mass formation rates are found from the difference F(ρ) − F(ρ + Δρ).

a. Tropics and subtropics

The equatorial ocean in the CO model on average receives about 30 W m⁻² more net heating than in the UN model and NCEP-STR climatology (Fig. 4), mostly in the form of enhanced solar flux (Fig. 5). The CCM3 radiative transfer equations do not account for the possible presence of excess solar absorption (Kiehl et al. 1998), leading to an increase over NCEP-STR in the tropical–subtropical insolation of 25–40 W m⁻² (see also Kiehl 1998). The solar flux is approximately balanced by elevated nonsolar heat losses in the subtropics (Fig. 5) due to a combination of increased wind speed relative to NCEP-STR, which is known to be too low in the Tropics (A. Dasilva 1997, personal communication), and a drier lower atmosphere (Hack et al. 1998). The net heat fluxes are approximately the same in the subtropical regions of the two models but with enhanced energy and hydrological cycles in the CO model (Figs. 3 and 4). The stratus cloud cover in the CO eastern boundary current upwelling systems is too low, resulting in even more solar heating, increased SSTs (2.0°–3.0°C), and large nonsolar heat flux biases.

The equatorial Pacific heating maximum, which was located near the South American coast in NCEP-STR, is shifted west in both model solutions and is narrower and stronger. Equatorial upwelling in the ocean model generates a pattern of cooler SSTs in the central Pacific, differences with STR about −2.0°C in the CO model and −1.0°C in the UN model, acting to reduce nonsolar heat loss in a region where recent data suggests that the STR SSTs may be too high (Gent et al. 1998).

The tropical CO net freshwater fluxes are dominated by an unrealistic, double annual mean ITCZ in both the Pacific and Atlantic, a weakly defined South Pacific subtropical convergence, increased precipitation over Indonesia, and a westward shift of the net deposition maximum from the eastern to western Indian Ocean Tropics (Fig. 7). The spatial gradients in the tropical F_net field are strong, and relatively small spatial displacements of P can lead to large local changes in the net freshwater flux of order ±2.0 m yr⁻¹, comparable in magnitude to the original signals in F_net. The CO tropical–subtropical SSSs are on average 0.5–1.0 psu lower than those of NCEP-STR at years 90–99, with the largest deviations under the southern limb of the ITCZ, in the Indonesian Archipelago and western equatorial Indian Ocean, and the SSS differences continue to increase through the CO integration.

The UN patterns of net freshwater deposition and evaporation are quite similar to the NCEP-STR patterns, but with a general reduction in magnitude of the ITCZ and subtropical evaporation regions due to a large extent to the salinity restoring term. The UN restoring term also crudely accounts for the missing riverine input, as marked by the larger net freshwater input near the mouth of major rivers (e.g., Amazon–Orinoco, Congo, Yangtze). An error signal from the absence of low-latitude rivers is less apparent in the CO model, with the exception of the Bay of Bengal, where the CO SSSs (not shown) are considerably larger than in the real ocean, which are kept low by the inflow from the Ganges–Brahmaputra.

The model and NCEP-STR climatological Indian–Pacific and Atlantic–Arctic water-mass transformations (Figs. 10 and 11) illustrate the typical patterns driven by basinscale air–sea fluxes (e.g., Speer and Tziperman 1995): net warming and freshening acting in concert to decrease the density of light, tropical surface waters; net evaporation in the subtropical density classes with a transition from net warming to cooling moving to higher density; and net cooling overwhelming the net freshwater input at subpolar latitudes tending to increase already large surface densities. The overall effect is to form either very light or very heavy surface waters at the expense of the intermediate surface density classes, which then must be replenished by interior mixing processes in the main thermocline (Tziperman 1986).

The increased tropical–subtropical warming, stronger hydrological cycle, and reduced SSSs in the CO model are evident in the water-mass transformations (Figs. 10 and 11) by an enhancement and shift toward lighter densities of the peak negative thermal and haline transformations. Compared with either NCEP-STR or the UN solution, the CO model produces a significant quantity of very light tropical surface water, 80 Sv in the Indian–Pacific basin at σ_θ 20.0–20.5 kg m⁻³ in the CO model versus about σ_θ 21.5 kg m⁻³ in both UN and NCEP-STR. By comparison, the UN simulation shows a general reduction in the amplitude of haline transformations relative to NCEP-STR due in part to the open-ocean salinity restoring term.

b. Midlatitudes and western boundary currents

The most prominent midlatitude signals in the model–data flux difference maps (Figs. 7 and 8) are found in regions of strong advection: the western boundary currents and the Antarctic Circumpolar current. Large negative net heat and freshwater fluxes are generated in the western boundary currents (e.g., Kuroshio, Gulf Stream, Agulhas) by the poleward transport of warm, saline subtropical surface waters, and substantial model–data discrepancies can arise when the strength and/or location of the model currents differs from the real ocean. Based on the barotropic mass transport streamfunction and SST fields, the model Gulf Stream and Kuroshio do not penetrate far enough into the ocean interior, the Agulhas retroflection is located too far east, and the Brazil–Malvinas confluence is translated equatorward relative to observations. Also, a portion of the model Kuroshio incorrectly flows through the Sea of Japan while the Gulf Stream separates from the coast too far north, resulting in the displacement of the cooler, fresh shelf waters along the Nova Scotia and Labrador coasts.

The resulting SST and net heat flux anomalies (Figs. 7 and 8) are qualitatively similar in the CO and UN solutions, underlining the fact that these problems appear to arise because of the expected poor representation of western boundary current physics in the coarse-resolution ocean model. The model responses differ considerably for the freshwater flux, however. In the CO solution, the large positive SST anomalies (2°–5°C) act to increase the evaporation rate leading to negative F_net anomalies. The elevated evaporation rates also occur in the UN case but are overwhelmed by the open-ocean salinity restoring term, which is responding to small positive SSS differences with climatology. This divergence in the two models is most apparent in the Kuroshio current and Gulf Stream, where several meters of excess nearshore freshwater deposition are observed relative to NCEP-STR. The midlatitude precipitation bands in the models are weaker than prescribed by the NCEP-STR climatology (Fig. 5), but this may have more to due with errors in stormtrack MSU data.

Considerable structure is observed in the model surface flux fields south and east of Africa in the Antarctic Circumpolar Current (ACC) associated with a train of warm/salty and cold/fresh surface anomalies. A region of particularly large net heat loss is found in both simulations downstream of the Kerguelan Plateau where the model ACC is deflected poleward by topography, and a band net annual heat loss extends farther east in the CO model along the ACC to south of Australia and New Zealand, collocated with a band of deep winter convection (Fig. 12). The validity of these features is difficult to judge at present with either the NCEP-STR or COADS climatologies because of the poor Southern Hemisphere data coverage.

The model Southern Ocean transformation diagrams (Fig. 13) show the formation of two distinct water masses, Antarctic Intermediate Water (CO σ_θ 25.5–26.5 kg m⁻³ and UN σ_θ 26.5–27.0 kg m⁻³) and Antarctic Bottom Water (CO σ_θ 28.0–28.3 kg m⁻³ and UN σ_θ 27.6–27.9 kg m⁻³). Both models contain a similar pattern with two cooling features in density space, one associated with the Agulhas and the ACC and the other with coastal deep water formation, with a band of zero or positive net heat flux in between. The freshwater flux is positive over most of the open Southern Ocean. The resulting convergence of F(ρ) supports the formation of about 50 and 35 Sv of Southern Ocean mode water and Antarctic Intermediate Water in the CO and UN model, respectively, consistent with the changes in the overturning circulation. The lower ACC region surface salinities in the CO model (Fig. 9) shift the intermediate water formation to considerably lighter densities and the bottom water to slightly higher densities, however, another symptom of the large-scale salinity drift in the CSM (Bryan 1997).

c. High latitudes, sea ice, and deep water formation

At high latitudes, the major signals in the model surface ocean flux fields are related to sea-ice formation/melting and deep water formation. The sites of model deep water formation are diagnosed from snapshots of ocean boundary layer depth (Fig. 13), a measure of the penetration depth for surface generated turbulence.

The net production, export, and subsequent melting of sea ice tends to create strong dipoles in the surface freshwater flux field, negative (positive) fluxes in ice formation (melt) regions (Figs. 4 and 6). The formation signature in Q_net is much weaker because the SSTs are already at or near the freezing temperature, and the heat losses to the atmosphere and ice are balanced by heat of fusion release. By comparison, the ocean supplies much of the heat required to melt sea ice, leading to strong cooling near the seasonal ice edge. For water-mass transformation, the overall effect of an active ice model is to separate (both in space and surface density class) the thermal and haline contributions arising from heat loss to the atmosphere in ice-covered regions.

Localized sea-ice formation rates, as large as 8–10 m yr⁻¹, can occur where offshore, wind-driven ice transport acts to maintain open ice leads in coastal regions. Such coastal ice production is observed in the CO model in the Southern Hemisphere in the Ross and Weddell Seas and along much of the Antarctic coast (Fig. 6). Sea ice is also produced near the northern edges of the Sea of Okhotsk and Bering Sea in the North Pacific, the Labrador Sea, and along the coasts of Spitsbergen, the Barents Sea, and Siberia in the Arctic. Sea ice is also formed in the open-ocean Arctic at the net annual rate of about 0.5 m yr⁻¹ (Fig. 4) in the CO solution. The corresponding primary meltwater signals appear as a rather zonally uniform band of freshwater deposition and cooling near 50°–60°S and as a sharp meltwater pulse at about 65°N east of Greenland due to ice export through the Denmark strait.

The CO model sea ice cycle is too strong relative to observations, in part because of the inadvertent use of an overly large ice drag coefficient (Weatherly et al. 1998), with the equivalent of over 3 m yr⁻¹ freshwater loss in the Antarctic zonal mean (Fig. 4). The resulting brine rejection drives sporadic coastal convection and persistent deep winter mixing in the eastern Ross and Weddell Seas (Fig. 13). Deep convection is also observed in the Arctic associated with the coastal ice formation regions. The North Atlantic sea-ice distribution extends too far south and east relative to observations, and the excessive meltwater input (Fig. 4) tends to suppress convection in the western subpolar gyre.

A broad region of net annual cooling and freshwater input is observed along the upper return path of the thermohaline circulation in the subpolar North Atlantic and Labrador Sea, with a minor Nordic Seas branch (2–3 Sv) (Gent et al. 1998) associated with the northward advection and cooling of Atlantic surface water in the Norwegian Sea. The CO production of North Atlantic Deep Water is about 25 Sv over the range σ_θ 27.6–28.1 kg m⁻³ (Fig. 11), with an additional 15 Sv of very dense water (σ_θ 28.5–28.7 kg m⁻³) from coastal and Arctic ice formation, much stronger than observational estimates. The Arctic solution is further degraded by the collapse of the Arctic halocline (Fig. 9) due to the absence of high-latitude river inflow (Boville and Gent 1998).

The UN formation rate is about 20 Sv with the deepest convection collocated with a region of large, negative net annual heat flux in the Labrador Sea. The overturning is close to observational estimates (Gent et al. 1997), but there is little Atlantic surface inflow or net water-mass formation in the Nordic Seas marked by too cold SSTs and weak cooling along the Norwegian coast. The strong under-ice restoring and zonal Fourier filtering combine to produce large-scale freshwater fluxes of alternating sign in the Arctic (Figs. 4 and 8) that, however, tend to cancel out over the basin mean (Fig. 13). The deep Arctic ventilation south of Spitsbergen is disconnected from the main thermohaline cell, and the large heat and freshwater fluxes along the east Greenland coast (Fig. 8) are associated with the strong under-ice restoring term and the model’s inability to represent the narrow outflow current of cold, fresh Arctic water.

The CO Southern Ocean surface waters are slightly cooler and significantly saltier than STR and Levitus (1982) along the coast and warmer (1.0°–2.5°) and fresher (0.5 psu) offshore (Fig. 9). The combination of more saline coastal water and stronger haline forcing generates nearly 15 Sv of dense Antarctic Bottom Water (σ_θ 28.0–28.2 kg m⁻³; Fig. 13), compared with 3–5 Sv in the real ocean, and is a major contributor to the CSM’s deep-ocean drift (Bryan 1998). Deep water formation in the UN model is isolated to the central Weddell Sea and to a lesser degree the Ross Sea and is supported by a negative net annual heat freshwater losses from the under-ice restoring terms. Only about 4 Sv of a less dense Antarctic Bottom Water (σ_θ 27.7–28.0 kg m⁻³) are formed in the UN model, closer to the observational estimates.

There is an indication of 1–2 Sv of intermediate water formation near 26.8–27.3σ_θ kg m⁻³ by haline processes in the CO North Pacific (Fig. 10). The central Bering Sea is a site of net cooling in both model solutions and the NCEP-STR climatology, but the cause appears to differ. The cooling region in the CO model is located in the central Bering Sea in the ice-melt zone, while a narrow, coastal feature is observed in the UN and NCEP-STR fields abutting the Siberian coast related to cold, dry surface air in the NCEP data.