A global atmosphere–ocean–sea ice general circulation model (GCM) is used in simulations of climate with present-day atmospheric CO2 concentrations, and with CO2 increasing to double the present-day values. The Parallel Climate Model includes the National Center for Atmospheric Research (NCAR) atmospheric GCM, the Los Alamos National Laboratory ocean GCM, and the Naval Postgraduate School dynamic–thermodynamic sea ice model. The ocean and sea ice grids are at substantially higher resolution than has been previously used in global climate models. The model is implemented on distributed, parallel computer architectures to make computation on the high-resolution grids feasible. The sea ice dynamics uses an elastic–viscous–plastic ice rheology with an explicit solution of the ice stress tensor, which has not previously been used in a coupled, global climate model.
The simulations of sea ice and the polar climate in the present-day experiment are compared with observed ice and climate data. The ice cover is too extensive in both hemispheres, leading to a large area of lower-than-observed surface temperatures. The Arctic exhibits a persistent high pressure system that drives the ice motion anticyclonically around the central Arctic. The ice thickness is greatest near the Chukchi Peninsula. Ice is exported through the Fram Strait, though the Canadian Archipelago, and inward through the Bering Strait. The modeled Antarctic sea ice moves at a faster speed than the observational data suggest. Many of the results and biases of the model are similar to those of the NCAR Climate System Model, which has the same atmospheric model component.
The response of the model to the increase in CO2 shows a significant thinning of the Arctic sea ice by 0.5 m but only a 10% decrease in ice area. Ice concentrations are reduced within the ice pack, while the ice edges are relatively unchanged. The Antarctic sea ice exhibits much less change in area and little change in thickness, in agreement with the reduced warming in the entire Southern Hemisphere.
Global coupled atmosphere–ocean–ice general circulation models (or “global climate models,” or GCMs) are some of the more comprehensive tools for investigating the large-scale behavior of the climate system, and for studying the role of the polar regions in global climate. They compute the balance of energy from a myriad of temporally varying influences, such as radiation, clouds, sea ice cover, ice motion, and atmospheric and oceanic heat transports. In recent years, coupled climate models have included more comprehensive, though sometimes crudely parameterized, treatments of the dynamics and thermodynamics of sea ice. Several GCMs (Pollard and Thompson 1994; Weatherly et al. 1998) have employed sea-ice dynamics using the cavitating-fluid assumptions of Flato and Hibler (1992). Ongoing developments in climate models, such as future versions of the National Center for Atmospheric Research’s (NCAR) Climate System Model (Boville and Gent 1998), are using viscous–plastic ice rheologies that represent the state of ice stress more realistically. The goal of such development is to represent accurately the role of sea-ice dynamics in its relationship to the other components of the climate system.
The simulation of polar climate in atmospheric general circulation models has numerous inherent difficulties associated with the convergence of grid points along meridians (cf. Randall et al. 1998). In addition, most global ocean models reside on Mercator (latitude–longitude) grids that have to treat the North Pole differently, and often use spatial filtering on the currents for numerical stability (cf. Gent et al. 1998). Sea-ice models on Mercator grids also have problems with the converging meridians near the pole (Weatherly et al. 1998). It would be advantageous to have the ocean and sea-ice models on grids that do not have a pole. The model in this study employs two such grids for ocean and sea ice, respectively.
Owing to computational costs, coupled climate models have not, in the past, been able to employ ocean grids of sufficiently high spatial resolution to represent either strong, narrow currents, such as the Gulf Stream, Kuroshio, and Equatorial Undercurrent, or narrow passages such as the Bering Strait, Canadian Archipelago, and the Strait of Gilbralter (e.g., Gent et al. 1998). The development of an efficient ocean model code for parallel computers (Dukowicz and Smith 1994) with a free surface formulation has made it feasible for higher-resolution ocean grids to be used in the coupled climate model presented here. A high-resolution grid for the dynamic sea-ice model is also presented here, which permits the inclusion of narrow straits in high latitudes, as well as permitting variability of sea-ice cover on more realistic spatial scales.
To make higher-resolution models computationally feasible, massively parallel processor (MPP) computers are increasingly being used at major computing centers, each with greater numbers of processors available. MPP computers also allow more complex physics packages to be included in models. The elastic–viscous–plastic dynamic ice rheology used in the sea-ice model presented here is one example of such physics. This study presents results from a global, coupled atmosphere–ocean–ice general circulation model in climate simulations with both present-day and increasing atmospheric CO2 concentrations. The focus of this paper is on the response of the sea ice and polar climate. Other aspects of these simulations, such as the global atmosphere and ocean are discussed in Washington et al. (2000).
2. Model description
This study uses the Parallel Climate Model (PCM), a coupled atmosphere–ocean–ice GCM developed through collaboration among NCAR, Los Alamos National Laboratory (LANL), and the Naval Postgraduate School, under the Department of Energy Climate Change Prediction Program. The PCM consists of the NCAR Community Climate Model version 3 (CCM3) atmospheric general circulation model (Kiehl et al. 1998) at T42 resolution and 18 vertical levels, the Parallel Ocean Program (POP) model developed at LANL (Dukowicz and Smith 1994), and the Naval Postgraduate School dynamic–thermodynamic sea ice model (Zhang et al. 1999). These components are linked by a component called the Flux Coupler that is based on the physics of the NCAR CSM Flux Coupler (Bryan et al. 1996), and modified for the PCM’s parallel architecture and the ice model physics (below). The land surface model (Bonan 1996) is also included as part of the atmospheric model CCM3. The PCM has some similarities to the NCAR CSM version 1.0 (Boville and Gent 1998) in that the CCM3 atmospheric GCM and land surface model are used, and the Flux Coupler physics are very similar. However, the PCM uses a different ocean component, POP, and a sea ice component with the elastic–viscous–plastic ice rheology, versus the cavitating-fluid ice model of CSM version 1.
The POP ocean is run on a dipole grid (see Washington et al. 2000), with the North Pole located in northern Canada and the South Pole remaining at 90°S. The grid has a global mean resolution of ⅔°, with the highest resolution of 25 to 60 km in the Arctic Ocean and Canadian Archipelago. The Bering Strait, Canadian Archipelago, and the Strait of Gibralter are resolved and open. The latitudinal spacing near the equator is reduced to ½°, which improves the simulation of El Niño phenomena associated with trapped equatorial Kelvin waves.
The sea-ice component of PCM from Zhang et al. (1999) solves for the evolution of the concentration, thickness, temperature, velocity, and snow depth in response to atmospheric and oceanic forcing. The ice dynamics uses the elastic–viscous–plastic (EVP) ice rheology of Hunke and Dukowicz (1997). This rheology explicitly solves the ice momentum equation using the ice stress tensor σi:
where m is the mass of ice, ui is the ice velocity, f is the Coriolis parameter, Hw is the dynamic sea surface height gradient τa is the atmospheric–ice stress, and τw is the ice–ocean stress. The EVP rheology includes an elastic term to the stress–strain relationship using a value for Young’s modulus E that is chosen to provide a stable solution for the desired numerical times step Δt. The efficiency of this formulation is that, for rigid ice with high bulk viscosity, the elastic term is greatest and the stress–strain equation resembles a damped wave equation, which has a stable solution for the explicit time-marching formulation. The explicit formulation is also relatively efficient on MPP architectures. The atmospheric stress on the ice is computed in the Flux Coupler using a stability-dependent formulation with a fixed aerodynamic ice roughness. The ice–ocean stress, τw, is computed in the ice model using the quadratic drag formula:
where Cw is the drag coefficient, ρw is the density of ice, uw is the geostrophic ocean current, and θw is a fixed turning angle (+25° in the Northern Hemisphere, −25° in the Southern Hemisphere).
The ice advection in PCM uses a modified Euler predictor–corrector scheme that has little diffusion, but is not conservative. Adjustments are made after advection to conserve ice mass and the total water content of the ice–ocean system.
The ice thermodynamics uses the “two-layer” version of the Semtner (1976) model with one internal ice temperature and one snow-layer temperature that approximates the thickness simulation in his three-layer model much better than in the zero-layer model. A single mean ice thickness and mean snow depth represent the ice and snow thickness in each grid cell, so multiple ice thickness categories are not represented. The surface temperature Ts is computed by solving the surface energy equation of Parkinson and Washington (1979) iteratively by the Newton–Raphson method. The surface albedo over sea ice is alb = 0.65 for frozen ice and snow and alb = 0.5 for melting ice and snow. The frozen ice/snow values are considerable lower than observed albedos (roughly 0.65 for bare ice to 0.85 for dry snow). The values were reduced in the model to adjust for the colder bias in polar surface temperatures, and were necessary for the adequate summer melting of snow from sea ice. The albedo of the ice-free ocean is computed with a zenith-angle dependence in the Flux Coupler. Where the ocean temperature falls below the freezing temperature (based on the local salinity), the ocean temperature is restored to freezing and the excess heat is passed to the ice model as the latent heat of ice formation. The ice formation from the ocean is assumed to occur in leads, and initially forms ice 0.50 m thick, which contributes to freezing over leads. Ice that is formed by conductive heat loss through the ice adds additional thickness onto existing ice. Stossel et al. (1996) showed that the method of parameterization of these ice processes has significant effects on climate model simulations.
The ice model is run on its own Cartesian grid of uniform 0.25° × 0.25° (27 km by 27 km) spacing, approximately equal to the highest grid spacing in the ocean grid in the Arctic. There are two separate domains over each polar region (Fig. 5). The western part of the North Pacific Ocean and the Seas of Okhotsk and Japan wrap around the right edge of the northern grid to the left edge. The ice model is run with a 6-h time step, and is coupled to the atmosphere and ocean every 1 day (the diurnal cycle of insolation is not seen by the ice model).
The interpolation of model variables and fluxes between atmosphere, ocean, and sea ice grids is performed in the Flux Coupler. An area-weighting method, developed at LANL for the POP model, is used to compute first-order interpolation weights for these differently oriented grids. Scaling the fluxes uniformly, so the hemispheric and global integrals are equal, ensures conservation during the interpolation between grids. However, this approach can present problems in areas like the Arctic, where the total ice-covered areas on the ice and ocean grids are not exactly identical because of their different orientations and resolutions.
Several observational and remotely sensed datasets have been used for comparison with the model results in this paper. Remotely sensed sea-ice concentrations calculated from Defense Meteorological Satellite Program’s Special Sensor Microwave Imager (SSM/I) over the period of 1979–91 [National Snow and Ice Data Center (NSIDC) 1997] were compared with the PCM ice concentrations. The monthly mean ice concentrations for these years were interpolated from their original 25 km to the 27-km PCM ice grid for comparisons to the model. Arctic buoy motions from the International Arctic Buoy Program (Rigor and Heiberg 1997) were used. For the Southern Hemisphere, the sea-ice motion (displacements) calculated from SSM/I brightness temperatures using maximum spatial correlations from Emery et al. (1997) were compared to model velocities, as was used in Weatherly et al. (1998). These satellite-derived ice motions were used in order to compare the large-scale velocity patterns, since the coverage of buoys for the Southern Hemisphere is limited. A comparison of satellite- and buoy-derived ice motions is presented in Kwok et al. (1998). Air temperature and sea level pressure data from the National Centers for Environmental Prediction (NCEP–NCAR) reanalysis were seasonally averaged over the period 1958–98 to compare with the model results.
4. Model experiments and results
The simulations presented in this paper include a“control” climate simulation, with an atmospheric CO2 concentration (pCO2) of 355 parts per million by volume (ppmv), and a simulation with pCO2 increasing by 1% per year until it reaches the doubled point at 710 ppmv at 70 yr and is held constant at 710 ppmv thereafter. Both simulations have been integrated for 300 model years. This paper will focus on the simulation of the polar climate in the PCM control simulation, and the response of the polar climate to doubling pCO2. Aspects of the global response to doubling pCO2 in the PCM, as well as the overall model description and initialization method, are described by Washington et al. (2000). The PCM experiments are initialized by integrating the coupled ocean and sea-ice components for 80 yr, using atmospheric forcing from a previous CCM3 simulation that uses observed sea surface temperatures and sea ice concentrations. The ocean spinup uses an acceleration factor of 10 in the deep ocean layers to bring the deep ocean in closer adjustment to the surface forcing. The surface temperatures and salinities are inequilibrium, with no significant trends present after 80 yr. The deep ocean does not reach equilibrium, although the trends decrease significantly by the end of the 80 yr.
The results shown in this paper are primarily 10-yr averages over the years 136 to 145 of each simulation. For the doubled-CO2 case, it begins 66 yr after the doubling point, and so the surface variables have reached new equilibrium values.
The seasonal mean surface air temperatures [December–January–February (DJF) and June–July–August (JJA)] from the PCM control are shown in Fig. 1, along with the temperature differences between the PCM control and NCEP–NCAR data, and between the control and the doubled-CO2 case. The central Arctic temperatures in DJF are −35° to −40°C, approximately 6°C colder than the observational data. The area of temperatures below 0°C covers the Greenland–Iceland–Norwegian (GIN) Seas, 12°C colder than observed, and coincident with too much sea-ice cover in PCM. The central Arctic temperatures in JJA are up to 4°C warmer than observed, although colder over the extensive southern ice-covered regions.
The temperatures over the Antarctic continent (Fig. 2) show a smoother gradient from the coast to the interior than observed, primarily attributable to the poor resolution of the topographic gradient of the Antarctic coast at this resolution of CCM3, which is approximately 300 km. In summer (DJF), the coastal temperatures in PCM are 2°–4°C too cold due to excessive sea-ice cover. In winter (JJA), the temperatures below 0°C extend farther equatorward due to excessive ice cover, but temperatures within the pack are warmer by up to 15°C due to areas of lower ice concentration created by the rapid divergence of the ice pack.
The global average surface air temperature in the doubled-CO2 case increases by 1.4°C (see Washington et al. 2000). There is a substantial hemispheric asymmetry in the warming; the Northern Hemisphere warms significantly more than does the Southern Hemisphere, where greater areas of ocean absorb the additional heating. The Arctic region’s (70°–90°N) annual average temperature increases by 3.6°C, more than twice the global average. In DJF the Arctic temperatures increase by 6.2°C, in JJA they increase by only 0.9°C (see Table 1), as they are mostly limited to 0°C at the melting sea-ice surface. Figure 1e shows that the greatest increase in DJF is 16°–18°C to the north of Iceland, where there is the greatest loss of sea ice, and 3°C warming in that region in JJA. There is actually a significant cooling of 1°–2°C in the North Atlantic southeast of Greenland, associated with a decrease in ocean heat transport to that area and the weakening of the meridional ocean circulation by approximately 0.25 × 1015 W (see Washington et al. 2000, Fig. 16). In the Southern Hemisphere (Figs. 2e,f), the air temperature increases from 1.5° to 3.0°C over Antarctica and the sea ice in both summer and winter.
The 10-yr mean sea level pressure (SLP) in the Arctic (Fig. 3) in DJF shows a high pressure of 1020 mb that extends from Siberia to northern Greenland, without the observed anticyclone in the Beaufort Sea, and geostrophic winds directed from the Kara Sea toward the Fram Strait. The PCM results show pressures 6–10 mb higher than the NCEP data over the Canada Basin of the Arctic Ocean, and 5 mb lower over the Norwegian Sea. In JJA the anticyclone is located near 85°N, 90°E with a central pressure of 1014 mb, in contrast with the more uniform observed pressures. This pressure pattern is nearly identical to that of the NCAR CCM3 simulations described in Briegleb and Bromwich (1998) and in the CSM by Weatherly et al. (1998). The dominant high pressure in the central Arctic and low pressure in the Norwegian Sea does not weaken sufficiently in summer, possibly as a result of the low-temperature bias. The persistent high pressure also tends to reinforce the low-temperature bias by limiting the inflow of warmer air from the south.
The Antarctic pressure pattern (Fig. 4) has a weaker gradient at the coast than observed, as expected from the resolution of the model, and exhibits the trough of low pressure of below 980 mb around most of the continent. Thus, a greater area of low pressure exists, 5–10 mb lower than observed in many places, similar to the CCM3 and CSM results mentioned above. This has the effect of increasing the geostrophic wind speeds around this low pressure, and contributes to the apparent overestimation of ice velocities simulated in the PCM.
The changes in SLP in the doubled-CO2 case are shown in Figs. 3e,f for the Arctic and Figs. 4e,f for the Antarctic. In DJF, the pressure decreases by 5 mb over Greenland, around Iceland, and over the Siberian coast of the Arctic Ocean. In JJA, the pressure decreases have a similar pattern but smaller magnitude, consistent with the smaller temperature change in summer. In general, the doubled-CO2 winds are weaker than those in the control run, without major shifts in pressure patterns. The Antarctic sea level pressure also decreases by up to 5 mb, mostly over the continent, which tends to weaken the polar pressure gradient and the circumpolar winds.
The seasonal and annual averages and net changes in the doubled-CO2 case in other surface variables and fluxes over the Arctic are listed in Table 1. The Antarctic changes are smaller than those in the Arctic and are not shown. Of particular interest is that one of the largest changes in the doubled-CO2 climate is the increase in specific humidity at the surface (+17% annually), total precipitation (+14%), and latent heat flux to the atmosphere (+18%, although still only 1.5 W m−2). This indicates a more vigorous hydrologic cycle in the Arctic, with increased moisture transport from the midlatitudes. The atmosphere contains and transports a greater quantity of water vapor, with precipitation increasing by 0.15 mm day−1 (+30%) in DJF, and resulting in 6-mm greater snow depth in DJF, although 6-mm less snow remains in JJA due to the greater heat fluxes. This additional water vapor results in greater cloudiness and decreased solar radiation at the surface (−5% annually), which also enhances the downward longwave radiation to the surface (+6%), which is a significant positive feedback to the warming.
b. Sea ice
The sea-ice concentrations in the control case in Fig. 5 for the Northern Hemisphere (NH) and Fig. 6 for the Southern Hemisphere clearly show that the sea-ice cover is too extensive in the PCM as compared with the satellite data. There is ice cover in PCM to the east of Greenland and north to Spitsbergen, the Barents Sea, the North Pacific, and the Gulf of Alaska, none of which should have significant ice cover. The August minimum ice cover is closer to the observed February ice, although there is still excessive ice in the Barents and Bering Seas. Ice concentrations are reduced to 0.7–0.8 in August in much of the Arctic, indicating that, while the ice edges extend too far south, there is substantial open water within the ice pack in summer.
In the doubled-CO2 case, the positions of the ice edges change remarkably little in winter, with more change in summer. The area of ocean that reaches the freezing point and maintains an ice cover in winter is relatively unchanged. However, the ice concentrations within the ice pack do change, particularly in summer, where they are reduced by 10%–15%. Figure 7 shows the total ice area in each hemisphere in the control and doubled-CO2 cases, as well as the total area computed from the satellite data interpolated to the PCM ice grid. The total NH ice area reduces by approximately 10%, and is still greater than the satellite data. The open water fraction, computed as the total area of open water within the ice pack divided by the total area contained by the ice edges, increases by about 25% in the Northern Hemisphere.
The ice concentrations in the Southern Hemisphere (SH) in Fig. 6 also show that the ice cover is too large in summer and winter. A large area of excess ice extends east from the Weddell Sea. This ice is driven from the Weddell Sea by the stronger wind forcing from CCM3’s pressure trough. In addition, the Antarctic Circumpolar Current (ACC) is significantly stronger in the PCM, being about 220 Sv (1 Sv = 1 × 106 m3 s−1), rather than the 120–150 Sv estimated from observations. The stronger ACC also tends to drive the ice eastward from the Weddell Sea. In comparison, the NCAR CSM simulations (Weatherly et al. 1998) also exhibited a strong ACC of 280 Sv, with strong transport of sea ice northward from Antarctica. However, the Southern Hemisphere ice cover in CSM remained very close to the observed ice cover in all months, in spite of the strong winds and ocean currents. In PCM, the SH ice cover changes very little in the doubled-CO2 case, in accordance with the reduction in the global warming signal in the SH in general.
The annual mean ice thickness for both hemispheres is shown in Fig. 8. The ice thickness in the control case is 2–3 m in the central Arctic; in summer, it is reduced to 2 m and less. It is about 0.5 m less than the ice thicknesses shown by Bourke and Garrett (1987) and from 1958 to 1976 by Rothrock et al. (1999), but closer to the thicknesses measured from 1993 to 1997. The spatial distribution of ice thickness does not exhibit the observed buildup of thick ice against northern Greenland and the Ellesmere Island associated with the Transpolar Drift Stream, as the PCM’s ice drift is shifted toward the Barents Sea. The thickest ice of 5–6 m is driven against the Chukchi Peninsula west of Bering Strait, and is 4–5 m between Greenland and Ellesmere Island (the Nares Strait) where the air temperatures are significantly lower than in the central Arctic. The Antarctic sea-ice thicknesses are mostly 0.5–2 m, the thickest ice being in the Weddell Sea along the Palmer Peninsula, and are not too different from observations. The doubled-CO2 case exhibits a significant thinning of Arctic sea ice of 0.5–1.5 m. The hemispheric average thicknesses in Fig. 9 show that the Arctic thickness difference is −0.48 m. This represents the biggest signal in the model’s polar regions in response to increased CO2. The ice thickness is reduced in response to higher air temperatures and increased longwave radiation, with somewhat reduced solar radiation. The slight increase in precipitation in the Arctic adds some insulating snow depth on top of sea ice that reduces thermodynamic ice growth in winter. The greater snow cover does not persist in summer (Table 1), so the additional precipitation may act as a positive feedback on the reduction of ice thickness in the doubled-CO2 case.
The Antarctic sea-ice thickness is reduced by 0.10 m in a band around the continent, and by 0.05 m overall, as the air temperature is higher by 0.5°C in these regions. In general, the warmer air occurs when there is no sea ice present, and the thermal inertia of the deep Southern Ocean is sufficient to absorb the additional heat without significant loss of ice cover.
The ice velocities for the control case are shown in Fig. 10. The Arctic ice velocities exhibit a central Arctic anticyclonic gyre centered on, and driven by, the dominant high pressure pattern. The Arctic ice speeds are comparable to those from observed buoy drifts in Fig. 10b, where the observed anticyclonic gyre (over years 1979–96) is located in the Beaufort Sea. There is significant flow of ice through Fram Strait, although this ice is thinner than the observed Fram Strait ice, since in PCM it originates in the Kara Sea and the drift stream is not across the pole. The monthly ice volume exports are shown in Fig. 11 for the Fram Strait, the Norway–Spitsbergen transect, the Canadian Archipelago, and the inflow (into the Arctic) through the Bering Strait. The Fram Strait export averages about 0.08 Sv, compared to observational estimates of 0.11 Sv (Aagaard and Carmack 1989), but has an appropriate seasonal cycle. The Norway transect is second largest at 0.02 Sv annually, so the total that enters the GIN Seas is 0.10 Sv, closer to the observed estimate. Also of interest is the much smaller export of ice through the Canadian Archipelago that has the reverse seasonal cycle because of the movement of ice in the summer, where the ice is mostly bound fast in these channels in winter. Figure 11a also shows the Fram Strait ice export in the doubled-CO2 case, which is reduced by about half due to the thinning of ice.
The Southern Hemisphere ice velocities in Fig. 10c exhibit features present in the satellite-derived velocity pattern, such as the coastal east wind drift and the cyclonic circulation in the Weddell Sea, as observed also from buoy drift patterns by Massom (1992). The PCM ice speeds are significantly faster than those from the satellite-derived data, noting the factor-of-four difference in vector scales. The version of SSM/I-derived drift speeds used here has recently been shown to underestimate the drift speed in the Antarctic by as much as half (and sometimes more), whereas little bias is seen in the Arctic (J. Maslanik 2000, personal communication). However, the modeled velocities still appear to large relative to the other datasets. As noted before, the larger ice speeds in PCM are driven in part by stronger winds from CCM3 (particularly the east wind drift adjacent to the Antarctic coast) and by faster ocean currents such as the ACC. The NCAR CSM simulations (Weatherly et al. 1998) also exhibited these large ice speeds, and were attributed in part to the use in CSM of a larger value of the aerodynamic roughness length (z0i) for sea ice than would commonly be observed for Antarctic sea ice. This PCM run (and more recent CSM runs) uses a lower z0i than was used in the first CSM runs, giving a bulk air–ice drag coefficient of Cd = 1.6 × 10−3. The lower air–ice drag can actually allow the surface winds to accelerate, producing a similar air-ice stress, resulting in similar ice speeds in both PCM and CSM. It is not known how much the ACC is forced by the ice–ocean surface drag and the net salinity flux from growing sea ice, and whether there is a mechanism that accelerates the ice–ocean system.
Although some aspects of the ocean simulation in the PCM runs are described in Washington et al. (2000), some analysis of the ocean heat budget may help to explain the PCM simulation of sea ice. The meridional global ocean heat transport (OHT) in the control PCM run (10-yr average) in petawatts (1 PW = 1015 W) is shown in Fig. 12, along with the observational estimate of OHT by Trenberth and Solomon (1994), and the implied OHT diagnosed from the surface energy fluxes from the uncoupled CCM3 simulation with prescribed, climatological monthly SSTs and sea ice concentrations (from SSM/I data) that is used to initiate the coupled PCM run. The implied OHT is the integral of the net surface heat flux and represents the meridional ocean heat transport that would be required to balance the net surface heat flux and maintain the present annual mean SSTs. Figure 12 shows that the implied OHT from the uncoupled CCM3 run is higher than the actual OHT in the PCM coupled run by approximately 0.25 PW at 65°N, and more negative by 0.20 PW at 65°S. Thus, when it is coupled to the PCM ocean–ice system, there is a net cooling at both 65°N and 65°S, coincident with the expansion of the ice cover. The coupled PCM’s actual OHT at 65°N is similar to that estimated by Trenberth and Solomon (1994), though they differ at other latitudes, and there is considerable uncertainty in the observational estimates.
The mean sea level rise in the Arctic in the doubled-CO2 case is listed in Table 1. This value for the thermal expansion has been computed by A. Craig (NCAR), using the temperature changes in each ocean column in PCM, as well as the local change in the free surface dynamic ocean level in POP (which averages globally to zero). The rise of 0.30 m is the result of a 0.11 m rise in the dynamic ocean level in the Arctic and a 0.19-m rise due to thermal expansion. The total 0.30 m is double the 0.15-m rise in the global sea level (due to thermal expansion only) predicted in this model. While the greater temperature change in the high-latitude ocean causes greater thermal expansion than the global average, the dynamic height change also contributes significantly to the total Arctic Ocean sea level rise. It also suggests that problems of coastal erosion attributable to sea level rise could reach higher levels in the Arctic earlier than they do in lower latitudes.
The largest notable bias in the polar regions in the control run of PCM is the excessive sea-ice cover in both hemispheres. The expansion of the ice cover is consistent with the imbalance between the ocean heat transport produced by the PCM ocean and that implied by the uncoupled CCM3 run. Some biases in the uncoupled CCM3 simulation in these regions have been analyzed by Briegleb and Bromwich (1998). They show a positive bias in the outgoing longwave radiation ofapproximately 10 W m−2 in the GIN Sea region in both DJF and JJA (their Figs. 3 and 4), and a deficit in absorbed shortwave radiation of 24 W m−2 in JJA and 9 W m−2 annually in the 70°–90°N region. A deficit in the longwave radiation reaching the surface of up to 20 W m−2 in the CCM3 radiative transfer model (RTM) has been diagnosed using atmospheric profile and radiation data from the Surface Heat Budget of the Arctic (R. Moritz 1999, personal communication). All of these biases act in the direction of a greater heat loss in the high northern latitudes, which can contribute to the larger implied OHT in CCM3 and the imbalance with the actual OHT in PCM. Planned improvements in the CCM3 RTM that account for longwave absorption in additional water vapor bands (based on the RTM of the Goddard Space Flight Center’s Data Assimilation Office) should make some positive correction to some of the radiative imbalance and cold bias in the CCM3 polar regions.
The unexpected result is that the extensive ice cover in the PCM control case does not rapidly disappear with the doubled-CO2 warming, although the Arctic ice thins by 0.5 m. Global climate model CO2 sensitivity studies by Pollard and Thompson (1994) and Rind et al. (1995;with dynamic ice models, but with mixed layer depth oceans) suggest that the greater the ice cover in the present-day model, the larger the apparent CO2 sensitivity, as more ice area disappears, particularly the initially thin ice. This does not appear to be true for PCM, where several processes may be acting to maintain the ice cover. The ocean heat transport in the North Atlantic decreases by about 0.25 PW between 40°N and 50°N in the doubled-CO2 case (Washington et al. 2000, Fig. 16), giving rise to colder SSTs that help to maintain the ice cover in the subpolar North Atlantic. The reduced snow albedo of 0.65 over sea ice in the control case decreases the albedo difference between snow and open ocean by 0.15, thus potentially reducing the ice-albedo feedback mechanism by 21%.
The thinning of sea ice in the doubled-CO2 run is also likely to be sensitive to the particular thermodynamic formulation, Curry et al. (1995) show that the Semtner three-layer model (which our PCM two-layer model closely approximates) had the largest (most sensitive) response of ice thickness to warming perturbations, while the Semtner zero-layer model was the least sensitive. The Curry et al. (1995) one-dimensional model with more detailed thermodynamics was less sensitive than the Semtner three-layer model, and its sensitivity was not increased dramatically by including explicit melt ponds or a 10-category ice thickness distribution model. Flato (1996) showed that the sensitivity of a two-dimensional ice model to a prescribed warming was reduced by the inclusion of open water leads and improved ice dynamics, as now included in the PCM. However, the two-dimensional model was more sensitive with the inclusion of an ice thickness distribution that allows thinner ice to disappear first (thus changing ice area). It appears that, in PCM, we have included a combination of processes that reduce the response in ice area (reduced ice-albedo feedback, dynamics, and leads) and enhance the response of ice thickness (the Semtner two-level model without an ice thickness distribution).
Another significant bias in the simulated Arctic climate in PCM is the persistent anticyclonic circulation centered near the pole. This greatly influences the ice velocity pattern and the ice thickness buildup on the Siberian side as opposed to the North American side. Higher-resolution atmospheric models may produce better Arctic surface pressure patterns, owing to their improved representation of topography (Greenland, Alaska, Ural Mountains). At least one simulation with CCM3 at T63 resolution has exhibited an improved Transpolar Drift Stream pressure pattern, but without a well-defined Beaufort Sea anticyclone. The improvements in the radiative transfer scheme in the CCM mentioned above also could increase the net surface heating in the Arctic and weaken the persistent high-pressure and low-temperature biases.
The ice velocities in PCM (and their influence on the thickness field) are also the product of the EVP ice rheology. It should be noted that identical control-climate simulations with the fully coupled PCM have been run with both the EVP rheology and viscous–plastic (VP) rheology of Zhang and Hibler (1997). The two rheologies produce nearly identical average velocity, concentration, and thickness fields, with only smaller local differences of velocities near the ice edge, where the EVP model tends to reduce the velocities. Comparisons of the VP and EVP dynamics by Hunke and Zhang (1999) also show similar results.
The simulations of sea ice and polar climate in the Parallel Climate Model experiments shown here exhibit many of the same biases as seen in the NCAR Climate System Model simulations (Weatherly et al. 1998). The extent of the sea ice is too large in the Northern Hemisphere, especially in the North Pacific. The buildup of ice thickness along the Siberian coast is driven predominantly by the high pressure in the central Arctic and the associated surface wind pattern. The Arctic surface temperatures are also lower than observations, which may also contribute to the dominant high pressure. The Antarctic sea-ice velocities are substantially larger than the satellite-derived average ice speeds and appear in conjunction with strong geostrophic winds and ocean currents around Antarctica. The similarities in these results support the hypothesis that the atmospheric forcing drives many of these patterns. The atmospheric component, CCM3, is the same in both models, while the ocean and sea-ice components are significantly different from CSM version 1.
The PCM also exhibits some marked improvements in the climate simulation over previous global, coupled climate models, although not particularly in the high-latitude atmosphere or sea ice. The resolution of narrow ocean currents such as the Gulf Stream, Kuroshio, the Equatorial Undercurrent, and around the continental shelves in the Arctic Ocean are significantly improved. In addition, the simulation of the temporal variability of El Niño is also well represented (Washington et al. 2000). These advances allow the PCM to be used to study further the connections between tropical and extratropical variability over interannual to centennial timescale. The high resolution of the sea-ice component allows for the explicit modeling of the Bering Sea, Fram Strait, and Canadian Archipelago ice transports, and the dynamic response to the eddies and boundary currents in the Arctic. However, the high-resolution sea ice does not noticeably improve the polar climate, because the atmosphere model resolution is still at 2.8°.
The simulation with increasing atmospheric CO2 in the PCM exhibits an interesting response in the polar regions. The primary response is the thinning of the Arctic sea ice an average of 0.5 m, with only a 10% reduction in the total Arctic area or extent. By comparison, Pollard and Thompson (1994) both show a decrease in fractional ice cover of about 28% in the Arctic and 46% in the Antarctic in their doubled-CO2 GCM experiment with dynamic sea ice and a mixed-layer ocean model. Rind et al. (1995) showed an approximate 30% decrease in global sea ice area and 60% decrease in ice thickness in their doubled-CO2 runs. The doubled-CO2 GCM experiment of Washington and Meehl (1996), which includes a global 1° resolution ocean model, shows a mostly ice-free Arctic in the summer only, with winter ice thinner by 1 m. The recently observed thinning of Arctic sea ice by Rothrock et al. (1999) is consistent with this aspect of the PCM doubled-CO2 simulation. However, it is not clear what has caused the observed sea-ice thinning or if it is the result of Arctic climate variability on multiyear timescales, and the PCM ice formulation is not comprehensive enough to represent changes in the ice thickness distribution accurately.
Because the present PCM ice model has a relatively simple thermodynamic formulation that may be biased toward thinning ice and little change in ice area, a new PCM sea-ice component is under development for future climate simulations. It incorporates the thermodynamic model of Bitz and Lipscomb (1999) and the ice thickness distribution model of Bitz et al. (2000), which should have substantial improvements over the current model.
The authors acknowledge the substantial contributions by the Parallel Climate Model effort headed by Dr. Warren Washington, including those of Tom Bettge, Tony Craig, Gary Strand, Vince Wayland, Jerry Meehl, Bert Semtner, Rodney James, Phil Jones, and Elizabeth Hunke. Jim Maslanik provided the satellite-derived ice data, based on data from the National Snow and Ice Data Center. This research is supported by the Climate Change Prediction Program of the U.S. Department of Energy and the National Science Foundation. The computer time is being provided by the NCAR Climate Simulation Laboratory, the DOE National Energy Research Scientific Computing Center, and the Los Alamos National Laboratory’s Advanced Computing Laboratory (ACL). The analysis and graphics in this paper were produced with Ferret.
Corresponding author address: Dr. John W. Weatherly, U.S. Army Cold Regions Research, 72 Lyme Road, Hanover, NH 03755.