In investigating the role of the polar regions in global climate, general circulation models (GCMs) are used to simulate the large-scale interactions between atmosphere, ocean, land, and sea ice. Such GCMs are also the most sophisticated tools available in estimating the sensitivity of the global climate to increasing greenhouse gases. Simulations of greenhouse warming are greatly affected by the ice-albedo feedback mechanism, in which reductions of sea-ice cover cause lower albedos and the polar amplification of the greenhouse warming (Budyko 1977). Thus, the simulation of polar climate in GCMs is a crucial part of understanding the behavior of the global climate system.
The difficulties inherent in modeling the polar climate with atmospheric GCMs are numerous, as summarized by Randall et al. (1998). The numerical problems of convergent grid points at the poles in spherical coordinates, the poor representation of water vapor and its transport (in spectral GCMs), the role of cloud microphysics, and the interactions between radiation, clouds, and surface albedo, all create biases in GCM simulations. Typical biases in the polar climate in GCMs are excessive precipitation and water vapor, cold upper-tropospheric temperatures, and poor seasonal cycles of cloudiness, pressure patterns, and longwave radiation (Walsh and Crane 1992; Battisti et al. 1992; Tzeng and Bromwich 1994).
Another major challenge in modeling the polar regions with a GCM is the representation of sea ice and its effects on climate. The ice cover limits energy exchanges between the atmosphere and ocean, but allows larger fluxes through open water (leads) within the ice pack. Ice surface albedo is a complex function of the variable physical state of the ice, the snow cover, and the presence of surface meltwater. The formation of sea ice rejects brine into the ocean, while melting ice returns freshwater to the ocean, which can influence global thermohaline circulation.
The impact of including more complete treatments of sea ice in atmospheric GCMs (with mixed layer or slab oceans) in determining the model sensitivity to increasing greenhouse gases has been investigated in several studies. Pollard and Thompson (1994) found that their climate model sensitivity was reduced by including sea-ice dynamics. Model sensitivity is also dependent on the mean ice thickness and ice extent achieved in the present-day climate simulations (Rind et al. 1995). Clearly, climate models need to produce realistic simulations of present-day sea ice and polar climate in order to more accurately estimate the sensitivity to climatic perturbations.
A small number of coupled atmosphere–ocean GCMs have included treatments of sea-ice dynamics, primarily for its potential effect on simulations of increasing greenhouse gases (Washington and Meehl 1996; Manabe et al. 1991). In order to limit the large computational requirements, these atmosphere–ocean GCMs still contain simplifications in important physical processes, such as the resistance of sea ice to shear stress, and computing radiation and energy fluxes through leads as well as over sea ice. As computational resources increase with the speed of supercomputers, it becomes possible to include in atmosphere–ocean GCMs sea-ice components that contain more complete and realistic parameterizations of physical processes.
Because of the significant impact of sea ice on climate, the Climate System Model (CSM), developed at the National Center for Atmospheric Research (NCAR), includes a sea-ice component that, although containing relatively simple dynamics and thermodynamics, still includes many physical processes that have not always been included in such models.
The purpose of this paper is to present some of the results in the polar regions of a 300-yr integration of the CSM, compare them with available analyses and observations, and to discuss reasons for major differences. Section 2 describes the CSM in general and the sea-ice component in particular. Section 3 describes the observational analyses used for evaluating the model results. Section 4 presents model results, compares them to observations, and presents several sensitivity tests of the sea-ice component. Section 5 summarizes the present model results and the ongoing model development.
2. Model description
The NCAR Climate System Model consists of the NCAR Community Climate Model version 3 (CCM3) atmospheric GCM at T42 resolution (2.8° × 2.8° grid, 18 vertical levels, Kiehl et al. 1996), a global ocean component with approximately 2° × 2° resolution and 45 vertical levels (NCAR Ocean Section 1996), and a dynamic–thermodynamic sea-ice component (Bettge et al. 1996). The model components are linked by the flux coupler (Bryan et al. 1996), which computes the interfacial fluxes between components. The sea-ice component will be described here; more complete descriptions of all components can be found in the references above, and an overall description of the CSM can be found in Boville and Gent (1998).
The sea-ice component solves for the dynamic and thermodynamic evolution of ice concentration, thickness, temperature, velocity, and snow cover in response to forcing by the atmosphere and ocean. The present version of the sea-ice component uses a global grid in spherical coordinates identical to that of the ocean component: fixed latitudinal spacing that varies from 2.2° at 20° latitude to 1.2° poleward of 60° latitude and longitudinal spacing of 2.4°. The southernmost row of grid cells are centered at 78°S, and there is an active grid cell that predicts thickness, temperature, and concentration at 90°N. Spitsbergen and Iceland are the only islands resolved in the northern high latitudes, as Greenland is included as a peninsula attached to Canada, and the Bering Strait is not resolved and is closed. The time step of the present component was initially set at 1200 s in the coupled integration, and was changed to 3600 s at year 60, with no substantial change in the results. The ice component responds to the diurnal cycle as resolved in the atmosphere.
The sea-ice thermodynamic formulation is based on the three-level Semtner (1976) model with minor modifications for its interface with the flux coupler. The flux coupler computes the atmospheric exchanges of heat, water, solar radiation, and momentum separately with the ice-covered and ice-free fractions of each grid cell, so that the effect of open leads on climate are included in CSM. Cooling the ocean temperature beyond the freezing point (primarily in leads) produces lateral ice growth (increasing the ice concentration), while heat loss via conduction through existing ice produces vertical growth (increasing thickness). The maximum ice concentrations are imposed to allow for the existence of leads in thick pack ice. Limits are 99% in the Northern Hemisphere and 96% in the Southern Hemisphere for ice thickness up to 1 m, and slowly increase for thickness >1 m following Harvey (1988). The surface albedo is computed as a weighted average of snow and bare ice albedos, using snow depth as a proxy for the fractional snow cover. The albedo of completely snow-free ice in CSM should be reduced to approximately 0.50 under melting conditions (Ts = 0°C); however, this was not implemented in the CSM experiment described here (C. Bitz 1997, personal communication). However, ice albedo is reduced to 0.50 when Ts = 0°C and partial snow cover exists, and the snow albedo is also reduced from 0.82 to 0.70. The net flux of freshwater from ice to ocean is computed from the net growth of ice and the melt of snow cover, using a constant salinity of 4 ppt for sea ice and 0 ppt for snow. All of the precipitation produced by the atmospheric component over sea ice falls as snow. If the ice surface temperature is 0°C, falling snow does not accumulate, and is input to the ocean as freshwater (CCM3 does not account for the latent heat of freezing in the formation of snow, so this change of phase does not require additional energy).
The momentum equation is solved by an iterative method based on FH92, which increments the internal ice pressure for grid cells under convergent stress, and recomputes the velocities from Eq. (1), until all the velocity magnitudes change by less than υcrit (=6 × 10−4 m s−1). For grid cells where the ice pressure P is less than the maximum yield strength Pmax (where Pmax is the function of thickness and concentration from FH92) the convergence of ice (−∇·ui) should be smaller than 10−11 s−1 (and any residual convergence of ice would be less than 0.1% per year). However, it was seen that there was a residual convergence of as much as 10−7 s−1 for points closest to the North Pole, even after as many as 1000 iterations, and with a much smaller υcrit = 10−6 m s−1. This residual convergence allowed ice thickness to increase at a rate of over 300% per year in a region where ice should be mostly incompressible. While this residual convergence is not significantly large on all spherical grids that have convergent grid points near the North Pole (G. Flato 1997, personal communication), it is a problem for the CSM grid, which has grid spacing of approximately 1 km at the northernmost velocity points.
Various datasets were used to evaluate the polar climate of CSM. Atmospheric sea level pressure from ECMWF analyses (Trenberth 1992) over the period 1980–89 are used. These analyses do include significant uncertainties in the polar regions due to both lack of data and operational-model limitations, but are one of few gridded global analyses available. The ECMWF TOGA Archive II (Trenberth 1992) annual data of precipitation minus evaporation (P − E) from the period 1985–92 were also used, which also contain significant uncertainty in the polar regions (Briegleb and Bromwich 1998).
Remotely sensed sea-ice concentrations calculated from Defense Meteorological Satellite Program (DMSP) Scanning Multichannel Microwave Radiometer (SMMR) and the Special Sensor Microwave/Imager (SSM/I) over the period 1979–91 were compared to CSM concentrations. For the Arctic basin, the mean ice motions of Colony and Thorndike (1984) obtained from drifting ice buoy data are used to compare to CSM ice velocities. Since no complete and long-term observations of buoy displacements are available for the Antarctic, remotely sensed ice motions represent the best available data for comparisons with the mean CSM fields. These data (Emery et al. 1998) consist of daily displacements of the sea-ice cover as analyzed from maximum spatial correlations in the observed 85-GHz brightness temperatures from SSM/I, mapped to a 62.5-km grid. Ice velocities for each grid cell represent the mean ice motion in that cell. The resulting ice velocities are closely parallel to the surface geostrophic wind fields and existing estimates of mean ice-drift patterns. The root-mean-squared (rms) difference between drifting buoys and the SSM/I-derived motions is approximately ∼7 km day−1 in the Antarctic. The mean difference is near zero, so temporal averaging serves to reduce the rms error. For this application, monthly mean ice velocities and ice concentrations were calculated from daily remotely sensed data for 1988–94. The SSM/I-derived ice concentrations and ice velocities were interpolated to the CSM ice grid by weighting the nearest grid points by the inverse of the square of their distance.
Since observational ice thickness data are limited in their spatial coverage and their frequency, particularly in the Southern Hemisphere, there is no purely observational estimate of ice volume transport. Therefore, an estimate of ice transport was derived using the monthly mean ice thicknesses from the CSM coupled integration (averaged over years 1983–97) and the monthly mean SSM/I ice displacements and concentrations from 1988 to 1994. Therefore, any biases in CSM thickness also affects the satellite-derived ice transport field. However, this allows us to compare the combined effects of different concentrations and velocities on total transport. The values of ice transport from CSM are computed using only mean monthly ice velocities, concentrations, and thickness. While this may miss variable transports caused by synoptic-scale storms in the model, the error in the monthly values of zonal mean transport are usually less than 10%, as compared with means computed from daily CSM output from individual years.
The coupled integration of CSM described here was conducted by the CSM Principal Investigators Group at NCAR. An overview of the experiment, including the method of initialization and spinup of the coupled model is described by Boville and Gent (1998). Starting from an equilibrium ocean run, the sea-ice and ocean components were spun up for 60 yr using atmospheric forcing from a previous simulation of CCM3. Then the fully coupled system was integrated for 300 yr from the end of that spinup period. This section presents the results from the atmosphere component in the polar regions at the surface, and from the sea-ice component. While the ocean also has a key role in determining the sea-ice cover, the results of the ocean component are not presented here, as they are discussed in other CSM-related papers (Gent et al. 1998; Bryan 1998; Danabasoglu 1998).
The atmospheric component results are averaged over an analysis period of 15 yr (years 1983–97) of the coupled integration. This period was chosen because the sea-ice coverage was typical of the coupled integration, and because 15 yr is of comparable length to most observational data and analyses. We also compared model data from years 183–197 and found similar atmospheric results to this period. The average ice velocities in all 10- and 15-yr periods of the 300-yr run did not indicate any noticeable variability that would suggest changes in surface winds, so this 15-yr period appears to be representative of the coupled run. It should be noted that these years do not correspond to any specific years of observational data, as they might if the model were forced with observed SSTs. The discussion here focuses on a few surface variables; more discussion of tropospheric results in the polar regions from an uncoupled integration of CCM3 are presented in Briegleb and Bromwich (1998).
The sea level pressure (SLP) field over each polar region is shown in Fig. 1. The annual SLP field in the Northern Hemisphere shows a high pressure center over the central Arctic, farther north and west than the observed Beaufort high. The annual cycle of the central Arctic pressure is much less in CSM than in the ECMWF analyses, with the December–February (DJF) high at 1016 mb compared to 1020 mb observed, and June–August (JJA) varying from 1012 to 1016 mb compared to the observed 1010 mb. Briegleb and Bromwich (1998) estimate the uncertainty of SLP at individual grid points in the ECMWF analyses to be ±2 mb. Differences between the 1980–89 analyses and 15-yr CSM means can also arise from the independence of time samples used, where longer datasets (e.g., 30–50 yr) might better represent present-day climate. In individual years, the location of the Arctic high in CSM appears to vary between the most frequent position shown in Fig. 1a and a less frequent position closer to the Siberian coast. The latter position produces surface winds and ice drift directed more toward the Bering Strait. The Icelandic low is slightly deeper (1001 mb) than in ECMWF analysis, although pressures over the Barents Sea (north and east of Norway) are closer, creating a stronger gradient between Iceland and Spitsbergen. The Aleutian Low is approximately as strong as observed, but is displaced to the north and it often has a separate low centered over the Sea of Okhotsk.
In the Southern Hemisphere, the Antarctic circumpolar trough has its minimum pressure positioned slightly north (at 62.8°S) of the observed trough (at 65°S), although this region may have the greatest uncertainty in ECMWF analyses due to sparse observations. Pressures over most of the ice-covered region are approximately 5 mb lower than observed. This SLP field is quite similar to that produced by the uncoupled CCM3 simulation, in spite of differences in the prescribed SST, sea-ice concentration and thickness in CCM3.
Surface temperatures in CSM (Fig. 2) in the central Arctic are typically −30° to −40°C in winter (DJF) over sea ice, which are on average 5°C colder than the temperatures observed from ice-station and buoy data (Martin and Munoz 1997), and up to 20°C colder over the East Siberian and Chukchi Seas (110°E–180°) where the CSM ice is several meters thick. Southern Hemisphere surface temperatures in CSM over sea ice (Fig. 2b) are −10° to −15°C near the continent. The CSM temperatures are about 5°C warmer than in the uncoupled CCM3 results because of the thinner ice and the fractional ice cover in CSM. This difference decreases with height to approximately 1°C at the 850-mb level, and is greatest (8°–10°C) over the winter sea ice. There are few observations of surface air temperature over sea ice in the Southern Hemisphere to compare to CSM, although temperatures along the Antarctic coast are colder than observations by about 10°C, similar to the uncoupled CCM3 results.
The values of precipitation minus evaporation are shown in Fig. 3 for the annual mean. The ECMWF analysis P − E fields are shown in Briegleb and Bromwich (1998), for which they estimate an uncertainty of ±10%. The average P − E over the Arctic (defined here as 70°–90°N) is 18.7 cm yr−1, compared to the value of 16.3 ± 0.5 cm yr−1 estimated from atmospheric moisture flux convergence (Serreze et al. 1995). Observed maxima on the western North American coast, on the southeast Greenland coast, and over the central Siberian plateau (90°E) are all simulated in CSM. Over the Southern Hemisphere ice pack, P − E also shows good agreement with observations in the magnitude and spatial pattern. The P − E maxima on the coastline at 50°E, 140°E, and 135°W occur downstream of the main trough centers, while the minima (over the Ross and Weddell Seas) occur where prevailing winds are off the continent. This pattern appears to be a marked improvement in the resolution of polar precipitation features over those of previous versions of CCM (cf. Walsh and Crane 1992). This improvement is primarily due to the increased horizontal resolution of CCM3 (to T42) and the use of semi-Lagrangian water vapor transport over the previous spectral transport (P. Rasch 1997, personal communication).
In summary, these atmospheric component results from CSM exhibit some areas of agreement with observations, and some biases. The patterns of SLP and P − E are most similar to the uncoupled CCM3 simulation, suggesting that these modeled fields are relatively insensitive to the differences in ice concentration and thickness in the two models, while the low-level air temperature does reflect the differences in ice conditions. In the next section, the response of the sea ice to this atmospheric forcing are presented, along with some model sensitivities that affect the CSM simulation.
b. Sea ice
The annual maximum and minimum ice areas in the Northern and Southern Hemispheres (NH and SH, respectively) are shown in Figs. 4a,b for the 60 yr of model spinup (left of year 0) and the 300 yr of the coupled integration. The maximum SH area during the spinup period is larger than observed, in response to the prescribed atmospheric forcing from a previous CCM3 run that did not distinguish fractional ice concentrations, thus creating colder temperatures over ice. The SH area remains close to the observed maximum and minimum areas throughout the coupled integration. The NH ice area is approximately 50% greater than the observed ice area at both maximum and minimum seasons through the coupled integration. The NH maximum area increases abruptly at year 105, when ice concentrations increase in the Labrador Sea, which then reverses around year 193.
The mean annual cycle of monthly ice areas and ice extents from CSM and from the satellite (SMMR and SSM/I) data in both hemispheres are shown in Fig. 5. Ice extent is defined as the sum of areas of grid cells where the concentration is greater than a threshold value (=0.15). The satellite values were interpolated to the CSM grid, so the Northern Hemisphere values are about 2 × 106 km2 less than those computed on the original SSM/I (25-km resolution) grid (as shown in Table 1), as the CSM grid masks some additional ocean points. The NH values from CSM are 30%–50% greater than the satellite values in all months. The SH ice extent from CSM is about 3 × 106 km2 greater than the satellite data in September, while the CSM ice area is up to 3 × 106 km2 less in October to November.
The interannual variability of the ice area, defined here as the annual mean of the standard deviation of the monthly ice area anomalies [σ(Ai)] over years 20–299, is compared in Table 1 to σ(Ai) from the satellite data (computed on the original SSM/I grid). Since only 13 yr of satellite data are used here, there is greater statistical uncertainty in σ(Ai) than from the 280 yr of CSM. The NH variability in CSM area is greater than observed, but constitutes the same fraction (4%) of the mean ice area as in the observations. The SH variability (9% of mean) is less than observed (12%).
The annual mean ice volumes, Fig. 4c, show the NH volume increases by about 50% in the first 20 yr of the coupled integration, in part because of an imbalance in the ice–ocean heat flux calculation, which was corrected at year 10. After year 10, there are decadal and multidecadal variations in the NH volume of up to 20%. These variations are attributable in part to the persistence of ice thickness anomalies in CSM similar to those modeled by Chapman et al. (1994). The SH volume in CSM shows mostly interannual variability, as the majority of the volume is the annual ice pack.
The area-averaged thickness (hi), in each hemisphere, defined as the area-weighted mean of ice thickness in the ice-covered fraction of each grid cell, has interannual variability similar to the total volume in Fig. 4c. The variability (Table 1) of the average thickness [(σhi)] in the Northern Hemisphere is 0.26 m, or 8% of the mean. Over just the Arctic basin, (hi) varies by 11% of the mean. While there is insufficient frequency of observations of Arctic sea-ice thickness to provide an accurate estimate of the interannual variability of the average thickness, models of Arctic sea ice provide proxy data that can be compared. The dynamic sea-ice–coupled ocean model by Weatherly and Walsh (1996) produces σ(hi) = 0.09 m (4% of its mean) over the entire Northern Hemisphere when forced by daily winds. The atmosphere–ice–ocean column model by Battisti et al. (1996) representing the central Arctic with a more comprehensive thermodynamic and albedo calculation, produces σ(hi) that is 27% of mean thickness. They also show that the simpler ice thermodynamics in the Manabe and Stouffer (1996) coupled GCM produces much less Arctic ice variability. The mean ice thickness in CSM is most likely sensitive to the snow-free melting-ice albedo (discussed later in this section). To determine whether the variability of ice thickness is sensitive to either the melting-ice albedo or ice dynamics would require additional coupled runs of significant length.
The mean ice concentrations from CSM from years 1983–97 and from the satellite observations are shown in Fig. 6. In February, CSM ice concentrations are primarily 96%–99% within the Arctic basin, while the ice cover in the Greenland–Iceland–Norwegian (GIN) Seas extends farther south and east than observed. The Labrador Sea is mostly ice free in this period, but becomes mostly ice-covered beginning at year 105, and returns to ice free by year 193. The cause of this change is difficult to diagnose in a fully coupled model, but it is associated with a decrease in the North Atlantic overturning, which reduces the northward ocean heat transport (A. Capotondi 1997, personal communication). The ice cover in the North Pacific Ocean is also too extensive and comprises the majority of the bias in total ice area. This region establishes a winter ice cover in the first few years of the coupled integration, due to excessive southward ice transport that cools the initially warmer SST. Ice concentrations in August also exhibit greater ice concentrations within the Arctic basin than observed, particularly over the continental shelf north of Siberia where open water is expected in summer:
In the Southern Hemisphere, CSM ice concentrations in August are similar to the observations in both extent and shape of the ice pack, except CSM exhibits a more gradual gradient at the ice edge, with lower concentrations within the ice pack. The highest concentrations occur in the western Weddell Sea, where the ice remains through the summer. The main discrepancies with the observations is in the Amundsen and Bellingshausen Seas, which lose their summer ice cover in CSM, and the eastern Ross Sea, which keeps its ice cover in summer.
The annual mean ice thickness in the Northern Hemisphere is shown in Fig. 7a. The Arctic thicknesses show a gradient from 1 m in the GIN Seas to a maximum of 12 m against the Bering Strait, which is closed in CSM. The model does not produce the observed pressure-ridged thickness buildup against North America (Fig. 7b). This pattern exists throughout the spinup and coupled simulation with the maximum thickness varying from 9 to 13 m. The ice thickness distribution in a dynamic ice model as in CSM is partially determined by the ice velocity field as the ice rheology responds to atmospheric and ocean forcing.
The annual mean ice velocities in the Northern Hemisphere in CSM are shown in Fig. 8a. The anticyclonic gyre at 150°W is consistent with higher pressures there, and with the thick ice at the Bering Strait that exerts an outward pressure. The pattern of transpolar drift originates from a more western position (90°–120°E) than the observed drift (Fig. 8b), also corresponding to the position of high pressure. The lack of shear strength in this ice rheology may prevent the buildup of ice thickness against northern Greenland where some shear is apparent. The drift pattern transports ice out of the Arctic (into the GIN Seas) with a mean ice volume transport of 0.15 Sv (Sv ≡ 106 m3 s−1) with a standard deviation (of annual means) of 0.025 Sv. This consists of ice transport through the Fram Strait (0.09 Sv) as well as across the Spitsbergen-to-Norway transect (0.06 Sv), although the latter path is not realistic for ice export. Aagaard and Carmack (1989) use an estimate of 0.10 Sv for ice export in Fram Strait (with substantial uncertainty). The ice exported through Fram Strait in CSM originates from farther west (45°–90°E) than the observed transpolar drift, and thus transports thinner ice than in reality, while the CSM velocities north of Fram Strait are considerably larger than the observed buoy-drift velocities in Fig. 8b.
The sensitivity of the Arctic ice thickness to different wind forcing was investigated by running the sea-ice component separately for several years forced by daily 10-m winds from the National Centers for Environmental Prediction (NCEP) reanalysis data. The other variables (air temperature, humidity, and incoming radiation) used the same daily CCM3 output (from a previous CCM3 run with prescribed SST and ice cover) asthe CSM model spinup. A slab ocean component that includes fixed monthly geostrophic currents was also used. Three experiments were run; a control using the CCM3 winds, and two using 1984–87 and 1988–91 NCEP winds, respectively. These experiments all used the same value of air–ice drag coefficient (Cd = 5.2 × 10−3) as the CSM run. The initial ice thickness in each run was the CSM ice thickness from 1 January, year 83. The ice thickness pattern in the control (CCM3 winds) was quite similar to the coupled CSM results (Fig. 7a) and are not shown. The mean thickness from February 1991 from the 1988–91 NCEP run is shown in Fig. 7b. It shows a gradient of thickness buildup against North America with a maximum of 9.8 m. Bering Strait still shows 8-m thickness, but the Siberian coast is dramatically thinner than in the annual mean in Fig. 7a. The 1988–91 winds indicate a much weaker Beaufort Sea anticyclone than CSM and transpolar winds directed toward Canada. The 1984–87 wind case (not shown) resulted in less overall change from the CCM3 case, as there was a stronger Beaufort Sea anticyclone in those years similar to CCM3 winds.
The sensitivity of Arctic ice thickness to the air–ice drag coefficient was also tested. The uncoupled model was forced with CCM3 winds and other variables as above, using a reduced drag coefficient of Cd = 1.6 × 10−3. With this change, the ice thickness in Bering Strait (not shown) was reduced to 5 m, with 3.5 m against the Chukchi coast at 150°E. This series of wind-forcing experiments show that the pattern of winds in CCM3 (and CSM) strongly influences the position of the ice thickness buildup, and the drag coefficient (and wind speeds) contributes directly to magnitude of the buildup. None of the model results show a maximum ice thickness buildup against Ellesmere Island and northern Greenland as observed (Fig. 8b). The fact that Bering Strait is closed in CSM also presents an unrealistic topographic barrier against which ice builds up.
Yet another sensitivity experiment with the uncoupled model tested the effect of including the melting-ice albedo for snow-free ice, as this effect was missing in the CSM run. The model was forced with CCM3 data and used the large drag coefficient. The mean Arctic ice thickness and total area was reduced by only 3%–5%, primarily in summer, while the Antarctic changed only negligibly. This actually reflects that the CCM3 forcing itself does not result in snow-free, melting ice with significant frequency. To correctly test this sensitivity requires an integration of the coupled atmosphere–ice system, so the air temperature–ice albedo feedback is included.
In the Southern Hemisphere, the annual mean ice thickness (shown in Fig. 7c) is primarily 0.25 to 0.5 m in the seasonal ice pack, with a buildup of 3 to 4 m against the Antarctic Peninsula. This is in general agreement with the pattern from the sparse observations, such as those of the Winter Weddell Gyre Study for the western Weddell Sea (Harder and Lemke 1994). The annual mean ice velocities in the Southern Hemisphere (Fig. 8c) show the westward drift along the East Antarctic Coast, the Weddell Sea gyre, and northeastward drift beginning around 65°S. The CSM ice speeds are several times greater than those analyzed from the SSM/I satellite data (noting the factor of 5 difference in scale between Figs. 8c and 8d) and have a greater meridional component. The observed ice velocities follow the isobars (Fig. 1d) in most regions, with some areas exhibiting an ice turning angle of 10° to 30° to the left, while the CSM velocities are predominantly oriented about 30° to the left of the CSM isobars (Fig. 1c), except along the Antarctic coast. The faster velocities in CSM are reflected in the meridional ice volume transport, shown in Fig. 9 for CSM and the values derived from satellite observations. The CSM transport is 0.11 Sv at the southernmost points (at 78°S), increasing to about 0.215 Sv at 74°S. This consists of northward transport in the Weddell and Ross Seas. The transports computed from the satellite data are near zero at 78°S and are 3 to 4 times smaller than the CSM values in general.
The large air–ice drag coefficient contributes directly to the faster velocities and large transport. The effect of applying a more realistic drag coefficient on ice transport is shown in Fig. 10. A sensitivity run consisting of the first 10 yr of the ice–ocean spinup was repeated with Cd = 1.6 × 10−3. Note that the transport values are significantly greater in the spinup than in the coupled run because of greater ice area and volume in the spinup (see Fig. 4). The sensitivity test shows the ice transport is reduced by approximately 50% for Cd = l.6 × 10−3, reflecting approximately 50% slower velocities. In a more recent CSM spinup that uses both the corrected air–ice drag and more recent CCM3 forcing data, the maximum ice transport in the SH is 0.09 Sv, considerably closer to the satellite-derived value. Another contribution to the faster velocities in the coupled CSM integration is the use of a linear drag law for the ice–ocean stress, which can underestimate the drag as velocities increase. While this affects the simulation for the large drag coefficient case, it is not as significant when the drag coefficient is reduced.
5. Discussion and conclusions
The sea-ice characteristics exhibited by the 300-yr CSM integration differ from observations in some noteworthy ways. In the Northern Hemisphere, the excess ice area accumulates within 20 yr in the North Pacific and North Atlantic Oceans. At its greatest extent (years 105 to 193) this buildup completely covers the Labrador Sea in winter; an unrealistic state for the modern climate. In the Arctic basin, the sea-ice extent is largely confined by the continents, but there is some contribution to the excess annual mean ice volume through larger-than-observed summer sea-ice concentrations. The CSM Arctic ice thickness distribution can be roughly described as a 90° westward rotation of the observed thicknesses about the pole. This pattern is strongly influenced by the CCM3 wind forcing, while the maximum ice thicknesses are created in part by the large air–ice drag coefficient.
The Southern Hemisphere sea-ice area is simulated better than the Northern Hemisphere. In particular, the annual cycle of area is generally bounded by the observed mean maximum and minimum, and the mean thicknesses are in general agreement with the limited observations. Sea-ice velocity in the CSM integration tends to be much faster than observed at all places and at all times, with some persistent bias in the direction of ice motion, caused for the greater part by the large air–ice drag coefficient. The faster ice velocities also exacerbate the numerical ice convergence/redistribution problem. Another serious consequence is the enhanced meridional ice transport away from Antarctica. The implied ice production poleward of 75°S (4 m yr−1), and hence the salt flux to the ocean due to brine rejection, are about 4 times greater than expected from observed ice velocities. The effects of the excess brine flux on the ocean are substantial (Doney et al. 1998).
Improvements to the ice–air and ice–ocean drag calculations are being incorporated into the flux coupler to reduce these biases. Furthermore, ways of overcoming the limitations of the present model’s ice dynamics are being explored. First, the cavitating-fluid dynamics formulation is undergoing further development aimed at correcting the residual convergence problem. Recent tests of the ice dynamics on a rotated, spherical grid with the equator placed over each pole show the residual convergence becomes negligible. Second, a new CSM sea-ice component is being developed that incorporates the Zhang and Hibler (1997) viscous plastic ice rheology, which includes shear viscosity. This model should provide a more realistic solution of ice velocities over that of the cavitating-fluid approximation, especially with the high-frequency coupling used in CSM (Lemke et al. 1998).
Despite the above imperfections in the sea-ice simulation, the CSM results show several significant improvements over previous coupled atmosphere–ocean–ice GCM simulations, for examples, the sea level pressure and precipitation–minus–evaporation in both polar regions, and the ice coverage and thickness in the Southern Hemisphere. These advances can be attributed in part to improvements in the atmospheric simulation in CCM3. The Southern Hemisphere ice cover, for example, is likely to be strongly dependent on the meridional energy transport in the atmosphere in southern latitudes. It is also dependent on the CSM ocean model, which apparently provides appropriate heating under and around sea ice, without recourse to flux adjustment.
The sea-ice component, and the CSM as a whole appears to have sufficiently complete processes to generate seasonal, interannual, and decadal variability, and to develop the negative feedbacks necessary to recover from significant and long-term perturbations to its climate state (e.g., the extensive sea ice of years 105 to 193). Battisti et al. (1997) show that natural variability in the Arctic is severely underestimated in the coupled GCM of Manabe and Stouffer (1996) because of its simplified ice thermodynamics. CSM includes more complete thermodynamics than the Manabe and Stouffer GCM and appears to have greater variability in Arctic thickness, as would be expected from the Battisti et al. (1997) analysis.
The small degree of climate drift in the 300-yr CSM integration is a major accomplishment, and is evident in the surface temperatures (Boville and Gent 1997), the sea ice (Fig. 4), and to some degree in the ocean (Bryan 1997). The present study suggests that this stability is robust, because it persists in the presence of significant variability. The study also demonstrates that small surface climate drift is achieved with a far from perfect sea-ice simulation. An encouraging conclusion is, therefore, that a considerable degree of component model imperfection can be tolerated along with realistic variability before surface flux corrections become necessary.
The authors would like to acknowledge Tom Bettge for his work on the sea-ice component, Brian Kauffman, Lawrence Buja, and the CSM principal investigators. This research was supported (for JWW) by the U.S. Department of Energy under the Carbon Dioxide Research Program, (for BPB) the National Science Foundation, and (for JAM) NASA Polar Oceans Grant NAGW-4402. Computing resources were provided by the Climate System Laboratory at NCAR.
Aagaard, K., and E. C. Carmack, 1989: The role of sea ice and other fresh water in the arctic circulation. J. Geophys. Res.,94, 14 485–14 498.
Battisti, D. S., D. L. Williamson, and R. E. Moritz, 1992: Simulation of the Arctic climatology with the NCAR CCM2. Preprints, Third Conf. on Polar Meteorology and Oceanogrography, Portland, OR, Amer. Meteor. Soc., 130–136.
——, C. M. Bitz, and R. E. Moritz, 1997: Do general circulation models underestimate the natural variability in the Arctic climate? J. Climate,10, 1909–1920.
Bettge, T. W., J. W. Weatherly, W. M. Washington, D. Pollard, B. P. Briegleb, and W. G. Strand, 1996: The NCAR CSM Sea Ice Model. NCAR Tech. Note NCAR/TN-425+STR, 25 pp. [Available from NCAR-CGD, P.O. Box 3000, Boulder, CO 80307.].
Bourke, R. H., and R. P. Garrett, 1987: Sea-ice thickness distribution in the Arctic Ocean. Cold Reg. Sci. Technol.,13, 259–280.
Boville, B., and P. Gent, 1998: The NCAR Climate System Model, version one. J. Climate,11, 1115–1130.
Briegleb, B. P., and D. L. Bromwich, 1998: Polar climate simulation of the NCAR CCM3. J. Climate,11, 1270–1286.
Bromwich, D. L., B. Chen, and R. Y. Tzeng, 1995: Arctic and Antarctic precipitation simulations produced by the NCAR community climate models. Ann. Glaciol.,21, 117–122.
Bryan, F. O., 1998: Climate drift in a multicentury integration of the NCAR Climate System Model. J. Climate,11, 1455–1471.
——, B. G. Kauffman, W. G. Large, and P. R. Gent, 1996: The NCAR CSM flux coupler. NCAR Tech. Note NCAR/TN-425+STR, 48 pp. [Available from NCAR-CGD, P.O. Box 3000, Boulder, CO 80307.].
Budyko, M. I., 1977: Climatic Changes. Amer. Geophys. Union, 261 pp.
Chapman, W. L., W. L. Welch, K. P. Bowman, J. Sacks, and J. E. Walsh, 1994: Arctic sea-ice variability: Model sensitivities and a multidecadal simulation. J. Geophys. Res.,99, 919–935.
Colony, R., and A. S. Thorndike, 1984: An estimate of the mean field of arctic sea-ice motion. J. Geophys. Res.,89 (C6), 10 623–10 629.
Danabasoglu, G., 1998: On the wind-driven ocean circulation of the uncoupled and coupled NCAR Climate System Ocean Model. J. Climate,11, 1442–1454.
Doney, S. C., W. G. Large, and F. O. Bryan, 1998: Surface ocean fluxes and water-mass transformation rates in the coupled NCAR Climate System Model. J. Climate,11, 1420–1441.
Emery, W. J., C. W. Fowler, and J. A. Maslanik, 1998: Satellite-derived maps of Arctic and Antarctic sea-ice motion: 1988–1994. Geophys. Res. Lett.,24, 897–900.
Flato, G. M., and W. D. Hibler III, 1992: Modeling pack ice as a cavitating fluid. J. Phys. Oceanogr.,22, 626–651.
Gent, P. R., F. O. Bryan, G. Danabasoglu, S. C. Doney, W. R. Holland, W. G. Large, and J. C. McWilliams, 1998: The NCAR Climate System Model global ocean component. J. Climate,11, 1287–1306.
Harder, M., and P. Lemke, 1994: Modeling the extent of sea-ice ridging in the Weddell Sea. The Polar Oceans and Their Role in Shaping the Global Environment, Geophys. Monogr., No. 85, Amer. Geophys. Union, 187–197.
Harvey, L. D. D., 1988: Development of a sea-ice model for use in zonally-averaged energy balance climate models. J. Climate,1, 1221–1238.
Jones, P. D., S. C. B. Raper, R. S. Bradley, H. F. Diaz, P. M. Kelley, and M. L. Wigley, 1986: Northern Hemisphere surface air temperature variations, 1851–1984. J. Climate Appl. Meteor.,25, 161–179.
Kiehl, J. T., J. J. Hack, G. B. Bonan, B. A. Boville, B. Briegleb, D. L. Williamson, and P. J. Rasch, 1996: Description of the NCAR Community Climate Model (CCM3). NCAR/TN-420+STR, 152 pp. [Available from NCAR-CGD, P.O. Box 3000, Boulder, CO 80307.].
Lemke, P., W. D. Hibler III, G. Flato, M. Harder, and M Kreysher, 1998: On the improvement of sea ice models for climate simulations: The sea ice model intercomparison project. Ann Glaciol.,25, 183–187.
Manabe, S., and R. Stouffer, 1996: Low-frequency variability of surface air temperature in a 1000-year integration of a coupled atmosphere–ocean–land model. J. Climate,9, 376–393.
——, ——, M. Spelman, and K. Bryan, 1991: Transient responses of a coupled ocean–atmosphere model to gradual changes of atmospheric CO2. J. Climate,4, 785–817.
Martin, S., and E. A. Munoz, 1997: Properties of the Arctic 2-m air temperature field for 1979 to the present derived from a new gridded dataset. J. Climate,10, 1428–1440.
McBean, G., 1986: The atmospheric boundary layer. The Geophysics of Sea Ice, N. Untersteiner, Ed., Plenum Press, 283–338.
NCAR Ocean Section, 1996: The NCAR CSM Ocean Model. NCAR Tech. Note NCAR/TN-423+STR., 88 pp. [Available from NCAR-CGD, P.O. Box 3000, Boulder, CO 80307.].
Pollard, D., and S. Thompson, 1994: Sea-ice dynamics and CO2 sensitivity in a global climate model. Atmos.–Ocean,32, 449–467.
Randall, D., J. and Coauthors, 1998: Status of and outlook for large-scale modeling of atmosphere–ice–ocean interactions in the Arctic. Bull. Amer. Meteor. Soc.,79, 197–219.
Rind, D., R. Healy, C. Parkinson, and D. Martinson, 1995: The role of sea ice in 2 × CO2 climate model sensitivity. Part I: the total influence of sea-ice thickness and extent. J. Climate,8, 449–463.
Semtner, A. J., 1976: A model for the thermodynamic growth of sea ice in numerical investigations of climate. J. Phys. Oceanogr.,6, 379–389.
Serreze, M. C., R. G. Barry, M. C. Rehder, and J. E. Walsh, 1995: Variability in atmospheric circulation and moisture flux over the Arctic. Philos. Trans. Roy. Soc. London,352A, 215–225.
Trenberth, K. E., 1992: Global analyses from ECMWF. NCAR Tech. Note NCAR/TN-373+STR., 191 pp. [Available from NCAR-CGD, P.O. Box 3000, Boulder, CO 80307.].
Tzeng, R. Y., and D. H. Bromwich, 1994: NCAR CCM simulation of present-day Arctic climate. Preprints, Sixth Conf. on Climate Variations, Nashville, TN, Amer. Meteor. Soc., 197–201.
Walsh, J. E., and R. G. Crane, 1992: A comparison of GCM simulations of Arctic climate. Geophys. Res. Lett.,19, 29–32.
Washington, W. M., and G. A. Meehl, 1996: High-latitude climate change in a global coupled ocean–atmosphere–sea-ice model with increasing atmospheric CO2. J. Geophys. Res.,101, 12 795–12 801.
Weatherly, J. W., and J. E. Walsh, 1996: The effects of precipitation and river runoff in a coupled ice-ocean model of the Arctic. Climate Dyn.,12, 785–798.
Zhang, J., and W. D. Hibler III, 1997: On an efficient numerical method for modeling sea-ice dynamics. J. Geophys. Res.,102, 8691–8702.
Annual means (
* An electronic supplement to this article may be found on the CD-ROM accompanying this issue or at http://www.ametsoc.org/AMS.
The National Center for Atmospheric Research is sponsored by the National Science Foundation.