• Anisimov, O., , B. Fitzharris, , J. O. Hagen, , R. Jefferies, , H. Marchant, , F. Nelson, , T. Prowse, , and D. G. Vaughan, 2001: Polar regions (Arctic and Antarctic). Climate Change 2001: Impacts, Adaptation and Vulnerability, J. J. McCarthy et al., Eds., Cambridge University Press, 801–841.

    • Search Google Scholar
    • Export Citation
  • Beckmann, A., , and R. Döscher, 1997: A method for improved representation of dense water spreading over topography in geopotential-coordinate models. J. Phys. Oceanogr, 27 , 581591.

    • Search Google Scholar
    • Export Citation
  • Boer, G. J., , G. Flato, , and D. Ramsden, 2000: A transient climate change simulation with greenhouse gas and aerosol forcing: Projected climate to the twenty-first century. Climate Dyn, 16 , 427450.

    • Search Google Scholar
    • Export Citation
  • Campin, J-M., , and H. Goosse, 1999: Parameterization of density-driven downsloping flow for a coarse-resolution ocean model in z-coordinate. Tellus, 51A , 412430.

    • Search Google Scholar
    • Export Citation
  • Chou, C., , and J. D. Neelin, 1996: Linearization of a longwave radiation scheme for intermediate tropical atmospheric models. J. Geophys. Res, 101D , 1512915145.

    • Search Google Scholar
    • Export Citation
  • Comiso, J. C., 2002: A rapidly declining perennial sea ice cover in the Arctic. Geophys. Res. Lett.,29, 1956, doi:10.1029/ 2002GL015650.

    • Search Google Scholar
    • Export Citation
  • Cubasch, U., and Coauthors, 1994: Monte Carlo climate change forecasts with a global coupled ocean–atmosphere model. Climate Dyn, 10 , 119.

    • Search Google Scholar
    • Export Citation
  • Cubasch, U., , G. C. Hegerl, , A. Hellbach, , H. Höch, , U. Mikolajewicz, , B. D. Santer, , and R. Voss, 1995: A climate change simulation starting from 1935. Climate Dyn, 11 , 7184.

    • Search Google Scholar
    • Export Citation
  • Cubasch, U., and Coauthors, 2001: Projections of future climate change. Climate Change 2001: The Scientific Basis, J. T. Houghton et al., Eds., Cambridge University Press, 525–582.

    • Search Google Scholar
    • Export Citation
  • Deleersnijder, E., , and J. M. Campin, 1995: On the computation of the barotropic mode of a freesurface world ocean model. Ann. Geophys, 13 , 675688.

    • Search Google Scholar
    • Export Citation
  • Deleersnijder, E., , J. P. van Ypersele, , and J. M. Campin, 1993: An orthogonal curvilinear coordinate system for a World Ocean model. Ocean Modelling, 100 , 710.

    • Search Google Scholar
    • Export Citation
  • Delworth, T. L., , R. J. Stouffer, , K. W. Dixon, , M. J. Spelman, , T. R. Knutson, , A. J. Broccoli, , P. J. Kushner, , and R. T. Wetherald, 2002: Review of simulations of climate variability and change with the GFDL R30 coupled climate model. Climate Dyn, 19 , 555574.

    • Search Google Scholar
    • Export Citation
  • Dickson, B., , I. Yashayaev, , J. Meincke, , B. Turrell, , S. Dye, , and J. Holfort, 2002: Rapid freshening of the deep North Atlantic Ocean over the past four decades. Nature, 416 , 832837.

    • Search Google Scholar
    • Export Citation
  • Dickson, R. R., , and J. Brown, 1994: The production of North Atlantic deep water: Sources, rates, and pathways. J. Geophys. Res, 99 , 1231912341.

    • Search Google Scholar
    • Export Citation
  • Dickson, R. R., , J. Lazier, , J. Meincke, , P. Rhines, , and J. Swift, 1996: Long-term coordinated changes in the convective activity of the North Atlantic. Progress in Oceanography, Vol. 38, Pergamon, 241–295.

    • Search Google Scholar
    • Export Citation
  • Dixon, K. W., , and J. R. Lanzante, 1999: Global mean surface air temperature and North Atlantic overturning in a suite of coupled GCM climate change experiments. Geophys. Res. Lett, 26 , 18851888.

    • Search Google Scholar
    • Export Citation
  • Fichefet, T., , and M. A. Morales Maqueda, 1997: Sensitivity of a global sea ice model to the treatment of ice thermodynamics and dynamics. J. Geophys. Res, 102 , 1260912646.

    • Search Google Scholar
    • Export Citation
  • Gent, P. R., 2001: Will the North Atlantic Ocean thermohaline circulation weaken during the 21st century? Geophys. Res. Lett, 28 , 10231026.

    • Search Google Scholar
    • Export Citation
  • Goosse, H., , and T. Fichefet, 1999: Importance of ice–ocean interactions for the global ocean circulation: A model study. J. Geophys. Res, 104 , 2333723355.

    • Search Google Scholar
    • Export Citation
  • Goosse, H., , and H. Renssen, 2001: A two-phase response of the Southern Ocean to an increase in greenhouse gas concentrations. Geophys. Res. Lett, 28 , 34693473.

    • Search Google Scholar
    • Export Citation
  • Goosse, H., , E. Deleersnijder, , T. Fichefet, , and M. H. England, 1999: Sensitivity of a global coupled ocean-sea ice model to the parameterization of vertical mixing. J. Geophys. Res, 104 , 1368113695.

    • Search Google Scholar
    • Export Citation
  • Goosse, H., , F. M. Selten, , R. J. Haarsma, , and J. D. Opsteegh, 2001: Decadal variability in high Northern latitudes as simulated by an intermediate-complexity climate model. Ann. Glaciol, 33 , 525532.

    • Search Google Scholar
    • Export Citation
  • Goosse, H., , F. M. Selten, , R. J. Haarsma, , and J. D. Opsteegh, 2002: A mechanism of decadal variability of the sea-ice volume in the Northern Hemisphere. Climate Dyn, 19 , 6183.

    • Search Google Scholar
    • Export Citation
  • Goosse, H., , F. M. Selten, , R. J. Haarsma, , and J. D. Opsteegh, 2003: Large sea-ice volume anomalies simulated in a coupled climate model. Climate Dyn.,20, 523– 536.

    • Search Google Scholar
    • Export Citation
  • Gordon, A. L., 1986: Interocean exchange of thermocline water. J. Geophys. Res, 91 , 50375046.

  • Gordon, H. B., , and S. P. O'Farrell, 1997: Transient climate change in the CSIRO coupled model with dynamic sea ice. Mon. Wea. Rev, 125 , 875907.

    • Search Google Scholar
    • Export Citation
  • Gregory, J. M., , R. J. Stouffer, , S. C. B. Raper, , P. A. Stott, , and N. A. Rayner, 2002: An observationally based estimate of the climate sensitivity. J. Climate, 15 , 31173121.

    • Search Google Scholar
    • Export Citation
  • Houghton, J. T., , Y. Ding, , D. J. Griggs, , M. Noguer, , P. J. van der Linden, , X. Dai, , K. Maskell, , and C. A. Johnson, Eds.,. 2001: Climate Change 2001: The Scientific Basis. Cambridge University Press, 881 pp.

    • Search Google Scholar
    • Export Citation
  • Huybrechts, P., , and J. de Wolde, 1999: The dynamic response of the Greenland and Antarctic ice sheets to multiple-century climatic warming. J. Climate, 12 , 21692188.

    • Search Google Scholar
    • Export Citation
  • IPCC Data Distribution Center, 1999: The IPCC data distribution center CD-ROM, Version 1.0. IPCC Secretariat, Hamburg, Germany. [Available online at http://ipcc-ddc.cru.uea.ac.uk/.].

  • Joos, F., , G-K. Plattner, , T. F. Stocker, , O. Marchal, , and A. Schmittner, 1999: Global warming and marine carbon cycle feedbacks on future atmospheric CO2. Science, 284 , 464467.

    • Search Google Scholar
    • Export Citation
  • Knutti, R., , and T. F. Stocker, 2002: Limited predictability of the future thermohaline circulation close to an instability threshold. J. Climate, 15 , 179186.

    • Search Google Scholar
    • Export Citation
  • Marshall, J., , and F. Schott, 1999: Open-ocean convection: Observations, theory, and models. Rev. Geophys, 37 , 164.

  • McCarthy, J. J., , O. F. Canziani, , N. A. Leary, , D. J. Dokken, , and K. S. White, Eds.,. 2001: Climate Change 2001: Impacts, Adaptation and Vulnerability. Cambridge University Press, 1032 pp.

  • Meehl, G. A., , W. M. Washington, , D. J. Erickson, , B. P. Briegleb, , and P. J. Jaumann, 1996: Climate change from increased CO2 and direct and indirect effects of sulfate aerosols. Geophys. Res. Lett, 23 , 37553758.

    • Search Google Scholar
    • Export Citation
  • Mellor, G. L., , and T. Yamada, 1982: Development of a turbulence closure model for geophysical fluid problems. Rev. Geophys. Space Phys, 20 , 851875.

    • Search Google Scholar
    • Export Citation
  • Miller, J. R., , and G. L. Russell, 2000: Projected impact of climate change on the freshwater and salt budgets of the Arctic Ocean by a global climate model. Geophys. Res. Lett, 27 , 11831186.

    • Search Google Scholar
    • Export Citation
  • Mitchell, J. F. B., , T. C. Johns, , M. Eagles, , W. J. Ingram, , and R. A. Davis, 1999: Towards the construction of climate change scenarios. Climatic Change, 41 , 547581.

    • Search Google Scholar
    • Export Citation
  • Nakićenovic, N., and Coauthors, 2000: Special Report on Emissions Scenarios. Cambridge University Press, 599 pp.

  • Nozawa, T., , S. Emori, , A. Numaguti, , Y. Tsushima, , T. Takemura, , T. Nakajima, , A. Abe-Ouchi, , and M. Kimoto, 2001: Projections of future climate change in the 21st century simulated by the CCSR/ NIES CGCM under the IPCC SRES scenarios. Present and Future of Modeling Global Environmental Change: Toward Integrated Modeling, T. Matsuno and H. Kida, Eds., Terra Scientific, 15–28.

    • Search Google Scholar
    • Export Citation
  • Opsteegh, J. D., , R. J. Haarsma, , F. M. Selten, , and A. Kattenberg, 1998: ECBILT: A dynamic alternative to mixed boundary conditions in ocean models. Tellus, 50A , 348367.

    • Search Google Scholar
    • Export Citation
  • Rahmstorf, S., 1995: Bifurcations of the Atlantic thermohaline circulation in response to changes in the hydrological cycle. Nature, 378 , 145149.

    • Search Google Scholar
    • Export Citation
  • Renssen, H., , H. Goosse, , T. Fichefet, , and J-M. Campin, 2001: The 8.2 kyr BP event simulated by a global atmosphere–sea-ice– ocean model. Geophys. Res. Lett, 28 , 15671570.

    • Search Google Scholar
    • Export Citation
  • Renssen, H., , H. Goosse, , and T. Fichefet, 2003: On the non-linear response of the ocean thermohaline circulation to global deforestation. Geophys. Res. Lett.,30, 1061, doi:10.1029/2002GL016155.

    • Search Google Scholar
    • Export Citation
  • Roeckner, E., , L. Bengtsson, , J. Feichter, , J. Lelieveld, , and H. Rodhe, 1998: Transient climate change simulations with a coupled atmosphere–ocean GCM including the tropospheric sulphur cycle. MPI Rep. 266, 48 pp.

    • Search Google Scholar
    • Export Citation
  • Roemmich, D., , and C. Wunsch, 1985: Two transatlantic sections: Meridional circulation and heat flux in the subtropical North Atlantic Ocean. Deep-Sea Res, 32 , 619664.

    • Search Google Scholar
    • Export Citation
  • Russell, G. L., , and D. Rind, 1999: Response to CO2 transient increase in the GISS coupled model: Regional coolings in a warming climate. J. Climate, 12 , 531539.

    • Search Google Scholar
    • Export Citation
  • Schaeffer, M., , F. Selten, , and R. van Dorland, 1998: Linking IMAGE and ECBilt. RIVM Rep. 481508008, 53 pp.

  • Schaeffer, M., , F. Selten, , J. D. Opsteegh, , and H. Goosse, 2002: Intrinsic limits to predictability of abrupt regional climate change in IPCC SRES scenarios. Geophys. Res. Lett.,29, 1767, doi:10.1029/ 2002GL015254.

    • Search Google Scholar
    • Export Citation
  • Schmittner, A., , M. Yoshimori, , and A. J. Weaver, 2002: Instability of glacial climate in a model of the ocean–atmosphere–cryosphere system. Science, 295 , 14891493.

    • Search Google Scholar
    • Export Citation
  • Schmitz Jr., W. J., 1995: On the interbasin-scale thermohaline circulation. Rev. Geophys, 33 , 151173.

  • Schott, F., , M. Visbeck, , and J. Fischer, 1993: Observations of vertical currents and convection in the central Greenland Sea during winter 1988/1989. J. Geophys. Res, 98 , 1440114421.

    • Search Google Scholar
    • Export Citation
  • Smethie, W. M. J., , H. G. Ostlund, , and H. H. Loosli, 1986: Ventilation of the deep Greenland and Norwegian seas: Evidence from krypton-85, tritium, carbon-14 and argon-39. Deep-Sea Res, 33 , 675703.

    • Search Google Scholar
    • Export Citation
  • Tartinville, B., , J. M. Campin, , T. Fichefet, , and H. Goosse, 2001: Realistic representation of the surface freshwater flux in an ice-ocean general circulation model. Ocean Modell, 3 , 95108.

    • Search Google Scholar
    • Export Citation
  • van der Wal, R. S. W., , and J. Oerlemans, 1994: An energy balance model for the Greenland Ice Sheet. Global Planet. Change, 9 , 115131.

    • Search Google Scholar
    • Export Citation
  • Visbeck, M., , J. Fischer, , and F. Schott, 1995: Preconditioning the Greenland Sea for deep convection: Ice formation and ice drift. J. Geophys. Res, 100 , 1848918502.

    • Search Google Scholar
    • Export Citation
  • Washington, W. M., and Coauthors, 2000: Parallel climate model (PCM) control and transient simulations. Climate Dyn, 16 , 755774.

  • Wood, R. A., , A. B. Keen, , J. F. B. Mitchell, , and J. M. Gregory, 1999: Changing spatial structure of the thermohaline circulation in response to atmospheric CO2 forcing in a climate model. Nature, 399 , 572575.

    • Search Google Scholar
    • Export Citation
  • View in gallery

    Dec–Jan–Feb (DJF) mean surface-air temperature averaged over the extratropical Northern Hemisphere (top lines, left axis) and over the GIN and Barents Seas (lower lines, right axis). A 21-yr running mean is applied to GIN and Barents Seas time series. The control simulation is indicated by a dotted line

  • View in gallery

    Vertical mixing depth (solid line, left axis) and surface-air temperature (dotted line, right axis) in Jan at 77°N, 12°E for one ensemble member in scenario A1b. A 3-yr running mean is used to filter out the strongest interannual variations, while still retaining the relevant time scales

  • View in gallery

    Change in annual-mean sea ice thickness (m; shaded) for 2071–2100 compared to 1961–90 for the ensemble mean. Contour lines indicate the change in convective mixing depth [contours (m) at … , −200, −100, +50]. The “extra” contour line indicates the division between areas where sea ice thickness decrease and increase

  • View in gallery

    DJF mean surface-air temperature change (°C) for 2071– 2100 compared to 1961–90 for the standard experiment ensemble mean

  • View in gallery

    Annual-mean surface-air temperature averaged over the GIN and Barents Seas for the relaxation experiment. The thick line indicates ensemble mean. For comparison, the ensemble mean of the standard experiment is also indicated (lower thick line)

  • View in gallery

    Three-year running mean of depth of convection (lower lines, left axis) at the location of maximum convection depth in the present-day model climatology. The upper lines show sea surface salinity in the grid point just outside and north of the area of present-day convection. Solid lines represent the standard experiment and dotted lines the Arctic sea surface salinity relaxation experiment

  • View in gallery

    Salinity for the ocean layer at 1500–3500 m for four consecutive 30-yr mean periods for one ensemble member in the standard experiment. The numbers in the 2071–2100 panel refer to the location of the points in the time series plot of Fig. 8

  • View in gallery

    Salinity in seven selected ocean grid cells in the layer at 1500–3500 m (left axis). The geographical location of the grid cells is indicated in the bottom-right panel of Fig. 7. Also shown in mean surface salinity in the Arctic Ocean (“arctic,” right axis)

  • View in gallery

    Semiequilibrium climate states characterized by long-term mean patterns of deep convection. Depth of convection (contour distance 100 m) is shown for 30-yr mean periods from (top) the 50% control run (state I) and from 250 yr after the freshwater pulse in (middle) the 30% experiment (state II) and (bottom) 10% experiment (state III)

  • View in gallery

    Annual mean GIN Sea deep-water formation for the standard 50% experiment and the 30% and 10% experiments. For convenience, the start of the historical spinups is labeled 1851, so that the freshwater pulse in the 30% experiment is labeled 1651

  • View in gallery

    Dec–Jan–Feb mean surface-air temperature change (°C) for 2071–2100 compared to 1961–90 for the 30% ensemble mean

  • View in gallery

    Change in annual mean sea ice thickness (shaded) for 2071–2100 compared to 1961–90 for the 30% ensemble mean. Contour lines indicate the change in convective mixing depth, with a contour distance of 100 m. The “extra” contour line indicates the division between areas where sea ice thickness decrease and increase

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 29 29 8
PDF Downloads 16 16 5

The Influence of Ocean Convection Patterns on High-Latitude Climate Projections

View More View Less
  • 1 Royal Netherlands Meteorological Institute (KNMI), De Bilt, Netherlands
  • 2 Université Catholique de Louvain, Louvain-la-Neuve, Belgium
© Get Permissions
Full access

Abstract

The mean state and variability of deep convection in the ocean influence the North Atlantic climate. Using an ensemble experiment with a coupled atmosphere–ocean–sea ice model, it is shown that cooling and subdued warming areas can occur over the North Atlantic Ocean and adjacent landmasses under global warming. Different “present-day” convection patterns in the Greenland–Iceland–Norway (GIN) Sea result in different future surface-air temperature changes. At higher latitudes, the more effective positive sea ice feedback increases the likelihood of changes in convection causing a regional cooling that is larger than the warming brought about by the enhanced greenhouse effect. The modeled freshening of deep ocean layers in the North Atlantic in a time period preceding a reorganization of GIN Sea convection is consistent with recent observations. Low-frequency internal variability in the ocean model has relatively little impact on the response patterns.

Current affiliation: National Institute of Public Health and the Environment (RIVM), Bilthoven, Netherlands

Corresponding author address: Michiel Schaeffer, National Institute of Public Health and the Environment (RIVM), PO Box 1, 3720 BA, Bilthoven, Netherlands. Email: Michiel.Schaeffer@rivm.nl

Abstract

The mean state and variability of deep convection in the ocean influence the North Atlantic climate. Using an ensemble experiment with a coupled atmosphere–ocean–sea ice model, it is shown that cooling and subdued warming areas can occur over the North Atlantic Ocean and adjacent landmasses under global warming. Different “present-day” convection patterns in the Greenland–Iceland–Norway (GIN) Sea result in different future surface-air temperature changes. At higher latitudes, the more effective positive sea ice feedback increases the likelihood of changes in convection causing a regional cooling that is larger than the warming brought about by the enhanced greenhouse effect. The modeled freshening of deep ocean layers in the North Atlantic in a time period preceding a reorganization of GIN Sea convection is consistent with recent observations. Low-frequency internal variability in the ocean model has relatively little impact on the response patterns.

Current affiliation: National Institute of Public Health and the Environment (RIVM), Bilthoven, Netherlands

Corresponding author address: Michiel Schaeffer, National Institute of Public Health and the Environment (RIVM), PO Box 1, 3720 BA, Bilthoven, Netherlands. Email: Michiel.Schaeffer@rivm.nl

1. Introduction

Using general circulation models (GCMs), projections have been made of significant future changes in climate as a result of increased concentrations of greenhouse gases (Cubasch et al. 2001). The largest differences between the projections of monthly mean surface-air temperature (SAT) change are found over the North Atlantic Ocean in (sub)polar areas (see, e.g., Fig. 9.10.a in Cubasch et al. 2001). The general warming over the North Atlantic is modified to show regions of subdued warming (Delworth et al. 2002; Gordon and O'Farrell 1997; Meehl et al. 1996; Roeckner et al. 1998), or even strong local cooling (Boer et al. 2000; Miller and Russell 2000; Mitchell et al. 1999; Nozawa et al. 2001; Russell and Rind 1999; Washington et al. 2000; Wood et al. 1999). The surface-air temperature response patterns over the North Atlantic, as well as their evolution in time, seem to be model dependent.

In this paper, we intend to shed light on the physical mechanisms that contribute to the different response patterns over the North Atlantic. The main regions of cooling as found in the model studies, which were mentioned earlier, are south of Greenland (Boer et al. 2000; Mitchell et al. 1999; Washington et al. 2000), south of Iceland (Miller and Russell 2000), or both (Nozawa et al. 2001), and in the Greenland–Iceland–Norway (GIN) Sea (Russell and Rind 1999; Wood et al. 1999). The climate changes over the North Atlantic are accompanied by changes in ocean convection and, in most cases, a weakening of the thermohaline circulation (THC). Most authors have explicitly linked local coolings to changes in convection, the exception being Washington et al. (2000), who link local cooling in the western North Atlantic to changes in mean atmospheric general circulation solely. In climate models, location and stability of present-day convection and sea ice cover depend on the formulation (parameterization) of physical processes. If response patterns are indeed sensitive to present-day location of convection, then model formulation plays an important role in explaining differences in response over the North Atlantic.

Next to the patterns of present-day convection, another cause of differences in model response over the North Atlantic is suggested when analyzing the results of Mitchell et al. (1999) and Cubasch et al. (1994). In both studies ensemble climate simulations with four members have been performed. Some members show local cooling in the North Atlantic in winter and some do not by the end of the simulation. Those differences between the individual members within an ensemble can only be explained by the model's internal climate variability. We have studied this potential influence of internal variability in the North Atlantic on the uncertainties in climate change projections in a recent paper (Schaeffer et al. 2002). We concluded that high-frequency internal variability (different initial conditions of the atmosphere) strongly limits the predictability in timing of a future sudden large-scale reorganization of convection patterns.

In the present paper we will focus on the sensitivity of North Atlantic climate change projections to differences in the present-day pattern of ocean convection. Different present-day convection patterns can result from differences in model formulation as mentioned earlier. However, such differences can also originate from the possibility of several equilibrium solutions of convection patterns for the same forcing. In the experiments presented in this paper, we have forced a coupled atmosphere–ocean–sea ice model (described in section 2) into various equilibrium states of mean large-scale convection patterns for present-day climate (section 3b). The different equilibrium states result in alternative climate projections over the North Atlantic in response to the same future forcing scenario. This allows us to examine the sensitivity of regional climate change projections in our climate model to the present-day location and intensity of convective activity in the North Atlantic (section 3c). Before addressing this central issue, we will go into some depth exploring the causes of the collapse of convection in our climate model. As sea ice plays a crucial role in air–sea exchange and changes in convection, we will examine the role of sea ice in triggering and amplifying changes in convection (section 3a). Finally, we will contrast the sensitivity of the response to the location and strength of convection with the internal low-frequency variability, by exploring the sensitivity to different “pre-industrial” initial conditions of the ocean model (section 3d). After addressing the issues mentioned earlier in section 3 we conclude with a short discussion in section 4.

2. Model description

The ocean–sea ice component CLIO (Goosse and Fichefet 1999) is made up of a primitive equation, free-surface ocean general circulation model (OGCM; Campin and Goosse 1999; Deleersnijder and Campin 1995) coupled to a comprehensive thermodynamic–dynamic sea ice model (Fichefet and Morales Maqueda 1997). The OGCM contains a detailed formulation of boundary layer mixing based on Mellor and Yamada's (1982) level-2.5 turbulence closure scheme (Goosse et al. 1999) and a parameterization of density-driven downslope flows (Campin and Goosse 1999). The horizontal eddy diffusivity and viscosity in the ocean are set equal to 300 m2 s−1 and 105 m2 s−1, respectively. Real freshwater fluxes (as opposed to negative salt fluxes) are applied at the ocean surface (Tartinville et al. 2001). The sea ice model takes into account the heat capacity of the snow ice system, the storage of latent heat in brine pockets trapped inside the ice, the effect of the subgrid-scale snow and ice thickness distributions on sea ice thermodynamics, the formation of snow ice under excessive snow loading and the existence of leads within the ice cover. Ice dynamics are calculated by assuming that sea ice behaves as a two-dimensional viscous-plastic continuum. The horizontal resolution is 3° × 3°. Vertically, there are 20 levels ranging in thickness from 10 m at the surface to 750 m in the deep ocean. A realistic bathymetry is used. The global grid is obtained by patching together two spherical grids. The first one is a standard spherical grid covering the whole World Ocean, except the northern Atlantic and the Arctic, which are represented on a spherical grid having its poles on the equator, in order to avoid the North Pole singularity. The two grids are connected in the equatorial Atlantic (Deleersnijder et al. 1993).

The atmosphere is represented by a global spectral quasigeostrophic model, truncated at T21 (ECBilt; Opsteegh et al. 1998), also referred to as an atmosphere model of “intermediate complexity,” with three vertical levels. ECBilt was developed for research on the relative importance of the physical feedbacks in the extratropics for decadal climate variability. ECBilt is further used for long simulations in paleoclimatological research and for different scenarios' projections of future anthropogenic climate change in ensemble mode. In the light of these research applications, ECBilt was developed with specific attention to high computational efficiency. Since we are mainly interested in the extratropics, we have adopted a quasigeostrophic approach for the dynamical core of the atmosphere model. In addition, physical parameterizations were kept as simple as possible. The physics package was recently extended with a computationally cheap radiation scheme (Schaeffer et al. 1998), developed by adapting the Greens' function approach (Chou and Neelin 1996).

The coupled model (ECBilt–CLIO) was recently used in a number of papers addressing the influence of high-latitude atmosphere–ocean–sea ice interactions on climate variability and change (Goosse et al. 2001, 2002; Renssen et al. 2001; Schaeffer et al. 2002). A surplus of precipitation with respect to observed rainfall is corrected by removing in each timestep 10% of rainfall for every grid cell in the North Atlantic Ocean and 50% for cells in the Arctic Ocean. The corresponding water is redistributed homogeneously over the North Pacific, a region where precipitation is underestimated in the model. Using this adjustment, a realistic present-day thermohaline circulation and seasonal cycle of Arctic sea ice margin are obtained. In this paper, we will modify the Arctic 50% part of the adjustment scheme to manipulate the strength and location of ocean deep convection. In the default ECBilt–CLIO configuration, the principal area of convection in the Northern Hemisphere ocean is located in the Greenland–Iceland–Norway Sea, south of Spitsbergen (Renssen et al. 2001). Strong convective activity in this region is confirmed by observational evidence (Dickson et al. 1996; Dickson and Brown 1994; Marshall and Schott 1999; Schott et al. 1993; Smethie et al. 1986; Visbeck et al. 1995). The present-day model climatology was explored in more detail by Goosse et al. (2001). They showed the principal mode of variability in the model, involving large variations in Arctic sea ice cover, to compare favorably with observations. The estimated global climate sensitivity is 1.7°C for a doubling of CO2, which is on the low end of the estimated range (Gregory et al. 2002; Houghton et al. 2001).

3. Collapse of GIN Sea convection in climate change ensemble experiments

ECBilt–CLIO was brought into equilibrium by running the model for 1000 yr with constant greenhouse gas concentrations of the year 1850. We then forced the model with transient historical greenhouse gas concentrations from 1850 up to 1960. From 1960 on, the runs continued in ensemble mode, first according to observed greenhouse gas concentrations up to 1990 and then by following the Intergovernmental Panel on Climate Change (IPCC) Special Report on Emissions Scenario (SRES) A1b scenario until year 2100 (Nakićenovic et al. 2000; Schaeffer et al. 2002), leading to a CO2-equivalent concentration of 1050 ppmv at the end of the simulation. We generated 10 ensemble members by applying insignificantly small random distortions in the atmospheric state in 1960. In addition, we have run a parallel “control simulation,” without any changes in greenhouse gases concentrations beyond the year 1850.

In Fig. 1, we show the transient development of Northern Hemisphere December–January–February (DJF) mean surface-air temperature. We see a consistent rise in temperature for the enhanced greenhouse experiment, which starts in the historical period, but only becomes clearly significant as compared to internal variability (noise) around the year 2000. Accompanying this rise in Northern Hemisphere SAT is a sudden strong cooling of SAT averaged over the GIN and Barents Seas (Fig. 1) around the year 2060. We will refer to this sudden cooling as the North Atlantic climate transition. The predictability of the timing of this transition is limited, as shown in Schaeffer et al. (2002). There is a difference of up to 40 yr in timing for the different ensemble members. Because of this predictability issue, all experiments presented in this paper were performed in ensemble mode. The individual simulations within one ensemble represent different evolutions of the climate system for the same external forcing, with equal probability of occurrence. The spread in the results among the individual ensemble members thus provides an indication of the influence of internal, or natural, climate variability on climate change projections.

In Fig. 2, we show for one ensemble member the surface-air temperature and the depth of convective mixing for the one grid cell that shows the strongest cooling in the greenhouse experiment. Surface-air temperature reacts promptly to the collapse of convection with a cooling of around 10°C. The local cooling is first caused by the reduction in the heat exchange between the deep ocean and surface waters, resulting from the shallower mixing. The heat capacity of the ocean mixed layer is reduced, so that cooling of surface waters and surface air in winter is more efficient. The cooling of the ocean's surface waters allows sea ice to expand southward in winter. Sea ice acts as an insulator and the added reduction in the exchange of heat between atmosphere and ocean contributes to a further drop in surface-air temperatures. At the same time, the reorganization of convection influences horizontal ocean currents and (poleward) heat transports.

The pattern of change in convection and sea ice coverage is shown in Fig. 3. Convection slows down near Spitsbergen, while other convection areas located southward, close to the coast of Norway, become somewhat more active. In observations (Comiso 2002) and model experiments (e.g., Anisimov et al. 2001; Cubasch et al. 2001), the area of Arctic sea ice is generally reduced under global warming, which also occurs in the experiments presented here. However, in these experiments, sea ice expands in the GIN Sea in the area of collapsed convection (Fig. 3). In Fig. 4, we show the pattern of surface-air temperature change for the greenhouse ensemble mean. The decrease in SAT occurs over a large area and coincides with the pattern of change for convective mixing depth and expansion of sea ice. The cooling extends over the North Atlantic and Arctic Oceans and causes cooling, or subdued warming over Scandinavia, Siberia, high-latitude North America, and Greenland. Note that although the major change in convection occurs in the GIN Sea (Fig. 3), we have included the SAT response in the Barents Sea in Fig. 1, because atmosphere and ocean transports cause SAT to respond equally strong in this area, as compared to the GIN Sea (see Fig. 4).

In the remainder of this paper, we will examine the causes and effects of this North Atlantic climate transition. The experiment just described will be referred to as the standard experiment.

a. Cause of the collapse of GIN Sea convection

1) Model sensitivity experiment

In the National Aeronautics and Space Administration Goddard Institute for Space Studies (NASA GISS) coupled GCM (CGCM) experiment of Russell and Rind (1999), a collapse of GIN Sea convection is attributed to increased poleward transport of sensible and latent heat in the atmosphere. The resulting ocean surface warming and freshening (decrease in evaporation minus precipitation) contributes to a stabilization of vertical stratification in the ocean. This eventually leads to a collapse of convection and regional cooling, comparable in amplitude and pattern to that in our experiments. They found thinning of sea ice in the Arctic, but did not consider this crucial for the explanation of the collapse of convection. Gordon and O'Farrell (1997), however, noted that in their CGCM global warming experiment ice melt contributed significantly to changes in the Arctic freshwater budget. In our experiments, thinning of sea ice contributes about a third of the total freshwater added to the ocean surface north of 60°N between 1900 and 2100. In the rest of this section, we will show that in our model, the Arctic freshwater budget plays a crucial role and that changes herein form a prerequisite for the collapse of convection.

We have repeated the standard experiment, while restoring surface salinity in the Arctic, north of about 80°N, in all ensemble members to climatological 1961– 90 annual-mean model values with a relaxation time scale of 2 yr (“relaxation experiment”). This time scale ensures that long-term salinity trends are removed, but seasonal cycle and variability on interannual time scales are filtered out only weakly. In this experiment Arctic sea ice is still allowed to react to global warming, taking into account the important high-latitude sea ice–albedo feedback, but, because Arctic sea surface salinity trends are filtered out, changes in Arctic sea ice no longer have a direct impact on the salinity budget of the GIN Sea.

In Fig. 5, we show for the 10 ensemble members surface-air temperatures averaged over the GIN and Barents Seas (cf. Fig. 1). Without the transport of low-salinity surface waters from the Arctic into the GIN sea, the probability of a collapse of convection and the resulting regional cooling is very low, if not zero. Thus, changes in atmospheric transport alone do not trigger a collapse of convection in our model. In Fig. 6, we show vertical mixing depth averaged over the same six grid points for one ensemble member from the relaxation and the standard experiment. Whereas vertical-mixing depth in the standard experiment collapses under influence of gradual freshening associated with changes in precipitation, evaporation, runoff, and Arctic sea ice thinning, this does not occur in the ensemble members of the relaxation experiment.

Goosse et al. (2001, 2002) have found a mode of decadal variability in ECBilt–CLIO, involving variations in Arctic sea ice volume. In this mode, a decrease in sea ice volume leads to the export of low-salinity anomalies to the GIN Sea and a (modest) reduction of convection. In a greenhouse scenario, the Arctic sea ice thinning and the associated reorganization of convection are much stronger. The GIN Sea response to Arctic sea ice thinning caused by global warming might be intertwined with this mode of decadal variability. Indeed, the internal variability has a strong influence on the timing of the collapse of convection in a greenhouse scenario, which leads to a limited predictability (see Fig. 1 and Schaeffer et al. 2002).

2) Comparison of modeled salinity trends with observations

We have shown that a collapse of GIN Sea convection in the model is caused by salinity changes. How do the simulated changes in salinity compare to observations as recently published by Dickson et al. (2002)? In the model, the salinity of ocean deep waters starts to decrease early on in the standard experiment around the location of strongest present-day convection (see Fig. 7 for salinity averaged over ocean model layers between 1500- and 3500-m depth). This freshwater anomaly gradually spreads throughout the GIN Sea. Eventually, low-salinity waters cross the Greenland–Scotland ridges and spread throughout the northwestern Atlantic. We have picked an ensemble member that showed a later collapse of convection, around the year 2060, than the ensemble member in Fig. 2, because now we focus on the period preceding the collapse. In Fig. 8, we have pictured the salinity time series of various selected points along a southward trajectory indicated in Fig. 7. Salinity trends in the points southward of the Greenland–Scotland ridge (points 2 to 7) seem to exhibit three stages. Initially, all points show a small downward salinity trend. These downward trends increase some years after the year 2030, the time around which the gradual decline of salinity in the deep waters of the GIN Sea spills over the Greenland–Scotland ridge (second stage). For points further down the trajectory, the increased trends are smaller, because of the more diffuse low-salinity “front” (see Fig. 7), but continue for a longer time period. The downward trends in all points are probably linked mainly to the down mixing and subsequent horizontal transport of surface salinity trends and not to the intensity of mixing, which shows no trend (see Fig. 2).

For point 1, located in the GIN Sea, the downward salinity trend continues until the end of the simulation. However, for points 2 to 4, southward of the Greenland– Scotland ridge, the final stage in the salinity time series is a stabilization, followed by a reversed upward trend near the end of the simulation. This reversed trend might be an indication of the changes in horizontal ocean transports, which are related to the collapse of GIN Sea convection.

The time series of grid cell 3 at about 60°N, 40°W near the southeastern tip of Greenland could be directly compared to near-bottom salinity measurements in the West Irminger Sea between 1965 and 2001 (Dickson et al. 2002). Whereas mean salinity is higher in our model (around 35.15, compared to around 34.9 psu in the measurements), the variability is comparable (see Fig. 2 in Dickson et al. 2002). The initial trends in this and other points are smaller in the model than the present-day observed trends of between −7 and −32 mpsu/10yrs at various measurement sites between 54° and 67°N (Dickson et al. 2002). After the spillover from the GIN Sea to the North Atlantic and preceding the collapse, the modeled salinity trends are equal to, or a factor of 2 larger than the observed trends. The best fit between modeled and observed trends is found when the second stage in the modeled salinity time series is used, roughly between 2020 and 2060, before the collapse of GIN Sea convection. However, as shown in Schaeffer et al. (2002), internal variability significantly reduces predictability of a sudden change in convection patterns. Combined with general modeling uncertainties, the timing of the associated changes in deep-ocean salinity is therefore uncertain.

Comparing the model results with observations leads to three suggestions. First, the observed downward trends in deep-water salinity in the North Atlantic might be associated with an earlier than modeled “spillover” of low-salinity deep-water anomalies from the GIN Sea, which is plausible given the uncertainty in timing. Since this spillover precedes a collapse of convection in the model, the observed trends may be a forerunner of a substantial reorganization of convection in the GIN Sea as a result of global warming, but are less likely to have been caused by such reorganization.

The second suggestion, deduced from the stronger observed present-day salinity trends, is that not all effects causing long-term salinity trends in observations are represented in the “present-day period” of the model experiment. Dickson et al. (2002) argue that upstream freshening in the GIN Sea has caused the observed salinity trends south of 67°N. This freshening may somehow be caused by the enhanced greenhouse effect, like in our model, but may also reflect the internal climate variability [e.g., the shift from low to high North Atlantic Oscillation (NAO) index during the measurement period].

Finally, we should add that measurements were performed over a relatively small depth range, in a well-defined current close to the bottom. Overflows are parameterized in the model (Beckmann and Döscher 1997; Campin and Goosse 1999), but even with this parameterization the simulated deep current is much more diffuse than the observed one, due to the coarse resolution and the difficulty to resolve well these overflows. The smaller trend in the model can be due, at least to some extent, to this excessive diffusion in the model.

b. Different equilibrium solutions of convection patterns for present-day climatology

In the stand-alone version of the ocean–sea ice model CLIO, which is forced by observed heat and water fluxes, the areas of deep convection in the North Atlantic are found in the Labrador and GIN Seas. In the coupled model, no convection occurs in the Labrador Sea. To maintain convection in the GIN Sea at a realistic high-latitude location, leading to a thermohaline circulation with a strength close to observations, a coarse-scale adjustment of precipitation is applied, as explained in section 2. In this section, we will show that it is possible to obtain in the model different present-day quasi-equilibrium climate states, characterized by different patterns of convection in the North Atlantic, by modifying those Arctic precipitation adjustments.

We show in Fig. 9 the present-day 30-yr mean depth of convective mixing for the standard experiment that uses a 50% adjustment of precipitation in the Arctic (climate state I). The maximum in the model's present-day climatology is found southwest of Spitsbergen. Main convection sites in the other solutions, referred to as states II and III, are located more to the south and are progressively weaker.

To force the model into these other quasi-equilibrium states, we have applied smaller Arctic precipitation adjustments (30%, 20%, 10%, and 0%). In each of the simulations of 150–250-yr length, GIN Sea convection weakened, but no large-scale reorganization occurred. The model was only forced into a different state when we subsequently applied a freshwater pulse to the area of maximum convection. This suggests that hysteresis (Rahmstorf 1995; Schmittner et al. 2002) is an important factor in the model's behavior. The pulses consisted of instantaneous changes of salinity in the upper 500 m in the 16 grid cells of maximum convection in the standard experiment, varying between −3 and −8 PSU as required to make the transition. Following the freshwater pulse, the area of maximum convection shifts to the southern edge of the GIN Sea in a new (quasi-) equilibrium state II (Fig. 9) and GIN Sea deep-water formation (DWF) drops from between 15 and 18 Sv (Sv = 106 m3 s−1) to around 10 Sv. This new state is only marginally stable for the 20%, 10%, and 0% adjustments. In these cases, the system made a spontaneous transition to a state III some 50 to 250 yr after the pulse, with GIN Sea DWF of around 5 Sv (see Fig. 10 for the results of the 10% adjustment).

In the model version with a 50% Arctic precipitation adjustment, a strong pulse decreased GIN Sea convection to the level associated with state II, but the system bounced back to state I after some 150 yr (not shown). For the 30% adjustment, the model stayed in climate state II for 600 simulation years following the freshwater pulse (Fig. 10). The additional equilibrium state II seems to be stable for a medium adjustment only. In the 30% experiment, it is stable on a time scale, which is long enough to repeat the greenhouse experiment starting from a different climate state than in the standard experiment.

Present-day North Atlantic maximum overturning in CGCMs used in climate change experiments ranges from 14.9 Sv (Dixon and Lanzante 1999) to 28 Sv (Gent 2001). Estimating it directly from observations is problematic, but values of around 15–20 Sv are mentioned in the literature (Gordon 1986; Roemmich and Wunsch 1985; Schmitz 1995). The present-day maximum overturning for both the 50% and 30% experiments, around 21 and 17 Sv, respectively, is consistent with these ranges, so that present-day climatologies of both experiments are acceptable in this respect. Sea ice boundaries are quite realistic for the 50% standard adjustment, but extend too much southward for the 30% adjustment experiment.

Direct measurements of deep convection in the GIN Sea are sparse. Smethie et al. (1986) infer from tracer studies that convection in the Greenland Sea must reach 3000 m, at least occasionally. Visbeck et al. (1995) found deep convection in the central Greenland Sea down to 1500 m in late winter of 1988/89, concluding that “convection was weak” in that season. Based on an extensive review of observations and reconstructions of convection in the Greenland and Labrador Seas, Dickson et al. (1996) found considerable variability in convection depth. Central Greenland Sea convection deepened from the late 1950s to early 1970s to 3500 m and then progressively weakened until convection reached no deeper than 1600 m in the 1980s. Thus, there is considerable evidence of deep convection in the GIN Sea. There is uncertainty in the principal locations, but most studies point to the central Greenland Sea near the principal area of convection in ECBilt–CLIO's 50% standard experiment. The main location of present-day convection in the 30% experiment is centered further south and is less realistic in this sense.

c. Sensitivity of climate change projections to present-day convection patterns

In this section, we will explore the influence of the location of present-day convection on climate projections. We will focus on northern mid- to high latitudes and on changes in ocean convection, ocean circulation, sea ice, and the effects on regional SAT change. To this end, we have performed a new ensemble scenario experiment, using as initial condition the model state at the end of simulation year 300 in the 30% experiment (for convenience labeled year 1850 in Fig. 10). For this 30% scenario experiment, the decrease of GIN Sea deep-water formation to around 5 Sv around the year 2050 is comparable to the “spontaneous” transitions from state II to III in the 20%, 10%, and 0% experiments (Fig. 10). In contrast, in the standard scenario experiment the collapse of convection resembles the transition from state I to II (Fig. 10).

In both the standard and 30% experiments, the increased convection south of the original convection site does not fully compensate for the reduction of GIN Sea deep-water formation to the north. Maximum North Atlantic overturning reduces by about 6 Sv (27% for the standard and 35% for the 30% experiment). This is close to the mean of the range of CGCMs in Houghton et al. (2001). North Atlantic deep-water export at 30°S, an indicator of the strength of the thermohaline circulation on a global scale, reduces by only 15% and 12%, for the standard and 30% experiments, respectively. This relatively mild response of the global thermohaline agrees with the results of Wood et al. (1999), who showed that convection areas in the GIN Sea can shift, with local cooling effects, without a major impact on the strength of the global THC, in their case none at all. Note that the sudden decrease in GIN Sea–wide SAT (Fig. 1) and the collapse of convection at the original convection site (Fig. 2) in the standard scenario experiment both occur after the year 2040. In contrast, the strength of GIN Sea DWF gradually starts to decrease already around the year 1970. This reduction in GIN Sea DWF is linked to the basinwide GIN Sea overturning, which shows a gradual slowdown that precedes the collapse of convection at the location of deepest convection by about 70 yr. This earlier reduction in GIN Sea DWF is probably related to the earlier stabilization of the ocean at other (shallower) convection sites, which apparently has a weak impact on GIN Sea mean SAT.

Compared to the standard experiment (Fig. 4) the ensemble mean pattern of SAT change is different for the 30% ensemble (Fig. 11). As for the standard experiment, we see an area of cooling, centered near the area of strongest deep convection in the present-day climatology (state II in this case), but the cooling signal is weaker compared to the standard experiment. One of the reasons is that deep convection in the present-day climatology is shallower and occurs in a smaller area (cf. states I and II in Fig. 9). The collapse of convection thus has less impact on the heat exchange between surface waters and the deep ocean.

The second, crucial factor in explaining the difference in amplitude is the different response of sea ice. The time period during which sea ice is present during winter, together with its thickness and surface area, determines the strength of the insulating effect of sea ice. Expansion of sea ice effectively means that local climate is transformed from marine to continental, with a stronger north–south temperature gradient and colder local winters. The changes of ice cover have a particularly large influence in the northern part of the North Atlantic, where the very strong oceanic heat transport warms the atmosphere. These strong heat transports are illustrated by the fact that ice-free areas are found even in winter at such high latitudes. By comparing Fig. 12 with Fig. 3, we see that the increase in sea ice thickness and area is much smaller in the 30% experiment. The lower temperatures at higher latitudes naturally allow for a stronger expansion of sea ice caused by the collapse of convection, and thus for a more effective insulating effect in the 50% experiment. Besides, the major changes in convection occur at about 65°N in the 30% experiment, a latitude where only a thin ice cover can exist, even in the absence of significant oceanic heat flux.

A comparison of the results of Russell and Rind (1999), showing strong cooling over the North Atlantic under global warming, with more recent experiments using a new version of the same GISS GCM shows that the cooling in the GIN Sea has been significantly reduced and shifted southward. Although many more changes were made in moving from one GISS model version to the next, based on our results we suggest that the difference in response patterns is strongly linked to the shift of major convective activity in the North Atlantic from the GIN Sea in the older model version to about 55°N, near the coast of Ireland (G. W. Russell 2001, personal communication) in the new model. This supports the conclusion that strong regional cooling depends on the location and strength of deep convection. The latter determines the location of maximum local cooling under global warming. In addition, the amplitude of the cooling increases with latitude. A higher latitude permits sea ice to exert a stronger positive feedback by its insulating effect.

d. Sensitivity of projections to initial conditions determined by low-frequency variability

In section 3c, we have shown that the present-day location of convection has a strong influence on high-latitude SAT response in climate change projections. In addition, we have addressed in an earlier paper the intrinsic limits to predictability of climate change projections, which result from the influence of high-frequency (up to decadal time scale) variability (Schaeffer et al. 2002). Here, we will assess the relative influence of different initial conditions in the ocean resulting from low-frequency internal variability.

Research on the “cold start problem” (e.g., Cubasch et al. 1995) indicated that initial conditions of the ocean, with respect to the model initially being in equilibrium or not, influences climate change projections. Present-day experiments generally start from equilibrium conditions for pre-industrial forcing to avoid this problem (e.g., Dixon and Lanzante 1999). However, in such experiments, the oceans' initial conditions are still potentially important, because the ocean circulation exhibits long memory relative to the time scale of anthropogenic climate change projections. As we know little about pre-industrial ocean conditions, it is relevant to investigate the sensitivity of climate change projections to this source of uncertainty.

We have repeated the standard 10-member ensemble integrations for initial conditions of the coupled model 300 yr later in the control simulation. This method is comparable to the ensemble integrations by Cubasch et al. (1994), Dixon and Lanzante (1999), and Mitchell et al. (1999). Differences between the two ensembles' present-day mean states are significant at the 95% confidence level in some regions, using Student's t test, especially in the North Atlantic, due to a different evolution of sea ice buildup in the Arctic. However, the difference between the mean responses of both ensembles to the scenario of future increase in greenhouse gas concentrations is only significant over Greenland. Likewise, we did not find a significant difference of the standard deviations of the response of both ensembles at the 95% confidence level (F-test). Note that this does not mean that mean and standard deviation of the climate response in the two ensembles are equal. Indeed, especially in the period of the GIN Sea convective regime transition, there are large differences between the two ensemble means, but, as variance within each ensemble is also large in the regions concerned, due to the limited predictability of the timing of the transition (see Fig. 1 and Schaeffer et al. 2002), these differences were found to be not significant. From this sensitivity experiment, we conclude that the results related to the strength and timing of regional cooling in the North Atlantic are robust for initial conditions of the ocean model, in as far as different initial conditions result from low-frequency internal variability. Note that we have only tested for two initial conditions, which were selected randomly.

4. Conclusions and discussion

Some of the largest differences between the responses of different CGCMs to increased greenhouse gas concentrations are found over the North Atlantic Ocean. As argued by Russell and Rind (1999), strong regional cooling in a warming climate can occur when convection and large-scale ocean circulation and thus horizontal heat transport change. In all CGCMs [except in Roeckner et al. (1998) and Gent (2001)] North Atlantic Deep Water Formation (NADWF) and the THC weaken under global warming. However, whether changes in convection lead to regional cooling depends on the balance between cooling by changes in ocean heat transport (advection and diffusion), cooling by reduction of ocean vertical mixing depth, changes in sea ice, local greenhouse warming, and changes in atmospheric heat transports.

Our results show that the location of deep vertical mixing in the ocean is crucial for the probability and strength of a regional cooling signal under global warming. A high-latitude location of convection holds the potential for a positive feedback involving expansion of sea ice. By its insulating effect, sea ice further stabilizes the stratification in the ocean and effectively shuts down convection. In addition, the insulating effect of sea ice implies that sea ice expansion at high latitudes is very effective in directly amplifying the cooling signal.

The timing of a sudden reorganization of convection has low predictability and influences the predictability of climate change projections (Knutti and Stocker 2002; Schaeffer et al. 2002). For example, of the four ensemble members of the greenhouse gas–only CGCM experiments of Mitchell et al. (1999), two ensemble members show no cooling in the North Atlantic. A third member shows a cooling starting in the 2020s in the Labrador Sea, which persists during the rest of the simulation, while the fourth member shows no cooling until this is introduced in the Labrador Sea in the 2080s (IPCC Data Distribution Center 1999). It would be interesting to see what happens when these integrations are continued, to see whether at some point cooling occurs in all ensemble members, as is the case in our model in the GIN Sea. The more southern location of present-day convection in the model of Mitchell et al. (1999) has the effect that the potential positive feedback involving sea ice is weaker.

Compared to current operational CGCMs, our climate model has a relatively low climate sensitivity. In a model with higher climate sensitivity, a stronger global warming might compensate for the regional cooling, thus resulting in subdued warming, rather than cooling areas. This would also have an influence on the feedback associated with sea ice. If the climate becomes warm enough, sea ice will not cover the sites where convection occurred previously, reducing the magnitude of the perturbation due to the modification of convection patterns. On the other hand, a higher climate sensitivity would lead to an earlier Arctic sea ice reduction and therefore an earlier shutdown of GIN Sea convection in a period where global warming is less pronounced, so that at least temporarily a strong regional cooling might occur.

This clearly illustrates that, in addition to the influence of decadal variability, the reorganization of the thermohaline circulation depends in a quite subtle way on the present-day climate simulated by the model and on its sensitivity to the increase in greenhouse gases. According to our results, it would thus be difficult to forecast the timing and the impact of such reorganization. However, the modeled deep-ocean salinity trends at a time period close to the collapse of GIN Sea convection compare well with present-day observations of Dickson et al. (2002). This suggests that observed salinity trends might be a forerunner of a substantial reorganization of GIN Sea convection.

Obviously, forecasting the impact and timing of events such as mentioned earlier, is further limited by model deficiencies. The ocean and sea ice models that we have used in this study are relatively advanced compared to the atmosphere model, which is of intermediate complexity. Compared to state-of-the-art CGCMs, our coupled climate model has lower spatial resolution and the atmosphere model uses a simplified physics package. Furthermore, alternatives exist for the vertical mixing and horizontal eddy diffusivity schemes used in the ocean component. However, we feel that the use of such alternative formulations is not crucial to the characteristics of the model as explored in this study (e.g., Goosse and Renssen 2001). The main purpose of our model is to generate and test hypotheses on climate (change) mechanisms that, ideally, would be further explored using more complex modeling tools. Our climate model compares well with observations and more complex models in terms of mean climate, climate variability, and atmosphere–ocean–cryosphere feedbacks, as shown in a range of recent papers (Goosse and Renssen 2001; Goosse et al. 2001, 2002, 2003; Renssen et al. 2003, 2001; Schaeffer et al. 2002). In our scenario experiments, changes in the Arctic freshwater budget decrease salinity in the GIN Sea, triggering a reduction of convection. The results of Russell and Rind (1999) indicate that a more complex atmosphere model coupled with somewhat less advanced ocean and sea ice components shows a strong cooling consistent with our model results over the GIN Sea, in response to a reduction of convective activity.

The North Atlantic regional cooling in our model extends over North Atlantic and Arctic Oceans and the Scandinavian, Siberian, high-latitude North American, and Greenland landmasses. We suggest that North Atlantic cooling might have possible consequences for climate, natural ecosystems, and human society in three ways. First, a strong cooling in the ocean mixed layer and associated changes in ocean circulation have a potential influence on marine life and the carbon cycle in this area (Joos et al. 1999). Changes in circulation and vertical mixing will influence nutrient availability and primary productivity, thus affecting, for example, the fisheries sector and the “biological pump,” which stimulates the absorption of atmospheric CO2 in the ocean. In addition, shallower mixing influences the “solubility pump.” The latter refers to the downmixing of cold surface waters and stimulates the uptake of CO2 from the atmosphere more effectively under conditions of deep mixing. A second category of impacts relates to the uncertainty in timing and extent of North Atlantic cooling. For example, since regional cooling moderates warming over Greenland, the uncertainties in North Atlantic cooling add to the uncertainty regarding Greenland ice sheet melting under global warming (Huybrechts and de Wolde 1999; van der Wal and Oerlemans 1994). A third category of impacts can be expected from the rate of change associated with the regional cooling in our study. Marine and terrestrial ecosystems, and human society, have some resilience with respect to internal climate variability (McCarthy et al. 2001). These considerations make it useful to further extend our knowledge of the interactions between different climate system components and carefully monitor the polar and subpolar regions in the decades to come.

REFERENCES

  • Anisimov, O., , B. Fitzharris, , J. O. Hagen, , R. Jefferies, , H. Marchant, , F. Nelson, , T. Prowse, , and D. G. Vaughan, 2001: Polar regions (Arctic and Antarctic). Climate Change 2001: Impacts, Adaptation and Vulnerability, J. J. McCarthy et al., Eds., Cambridge University Press, 801–841.

    • Search Google Scholar
    • Export Citation
  • Beckmann, A., , and R. Döscher, 1997: A method for improved representation of dense water spreading over topography in geopotential-coordinate models. J. Phys. Oceanogr, 27 , 581591.

    • Search Google Scholar
    • Export Citation
  • Boer, G. J., , G. Flato, , and D. Ramsden, 2000: A transient climate change simulation with greenhouse gas and aerosol forcing: Projected climate to the twenty-first century. Climate Dyn, 16 , 427450.

    • Search Google Scholar
    • Export Citation
  • Campin, J-M., , and H. Goosse, 1999: Parameterization of density-driven downsloping flow for a coarse-resolution ocean model in z-coordinate. Tellus, 51A , 412430.

    • Search Google Scholar
    • Export Citation
  • Chou, C., , and J. D. Neelin, 1996: Linearization of a longwave radiation scheme for intermediate tropical atmospheric models. J. Geophys. Res, 101D , 1512915145.

    • Search Google Scholar
    • Export Citation
  • Comiso, J. C., 2002: A rapidly declining perennial sea ice cover in the Arctic. Geophys. Res. Lett.,29, 1956, doi:10.1029/ 2002GL015650.

    • Search Google Scholar
    • Export Citation
  • Cubasch, U., and Coauthors, 1994: Monte Carlo climate change forecasts with a global coupled ocean–atmosphere model. Climate Dyn, 10 , 119.

    • Search Google Scholar
    • Export Citation
  • Cubasch, U., , G. C. Hegerl, , A. Hellbach, , H. Höch, , U. Mikolajewicz, , B. D. Santer, , and R. Voss, 1995: A climate change simulation starting from 1935. Climate Dyn, 11 , 7184.

    • Search Google Scholar
    • Export Citation
  • Cubasch, U., and Coauthors, 2001: Projections of future climate change. Climate Change 2001: The Scientific Basis, J. T. Houghton et al., Eds., Cambridge University Press, 525–582.

    • Search Google Scholar
    • Export Citation
  • Deleersnijder, E., , and J. M. Campin, 1995: On the computation of the barotropic mode of a freesurface world ocean model. Ann. Geophys, 13 , 675688.

    • Search Google Scholar
    • Export Citation
  • Deleersnijder, E., , J. P. van Ypersele, , and J. M. Campin, 1993: An orthogonal curvilinear coordinate system for a World Ocean model. Ocean Modelling, 100 , 710.

    • Search Google Scholar
    • Export Citation
  • Delworth, T. L., , R. J. Stouffer, , K. W. Dixon, , M. J. Spelman, , T. R. Knutson, , A. J. Broccoli, , P. J. Kushner, , and R. T. Wetherald, 2002: Review of simulations of climate variability and change with the GFDL R30 coupled climate model. Climate Dyn, 19 , 555574.

    • Search Google Scholar
    • Export Citation
  • Dickson, B., , I. Yashayaev, , J. Meincke, , B. Turrell, , S. Dye, , and J. Holfort, 2002: Rapid freshening of the deep North Atlantic Ocean over the past four decades. Nature, 416 , 832837.

    • Search Google Scholar
    • Export Citation
  • Dickson, R. R., , and J. Brown, 1994: The production of North Atlantic deep water: Sources, rates, and pathways. J. Geophys. Res, 99 , 1231912341.

    • Search Google Scholar
    • Export Citation
  • Dickson, R. R., , J. Lazier, , J. Meincke, , P. Rhines, , and J. Swift, 1996: Long-term coordinated changes in the convective activity of the North Atlantic. Progress in Oceanography, Vol. 38, Pergamon, 241–295.

    • Search Google Scholar
    • Export Citation
  • Dixon, K. W., , and J. R. Lanzante, 1999: Global mean surface air temperature and North Atlantic overturning in a suite of coupled GCM climate change experiments. Geophys. Res. Lett, 26 , 18851888.

    • Search Google Scholar
    • Export Citation
  • Fichefet, T., , and M. A. Morales Maqueda, 1997: Sensitivity of a global sea ice model to the treatment of ice thermodynamics and dynamics. J. Geophys. Res, 102 , 1260912646.

    • Search Google Scholar
    • Export Citation
  • Gent, P. R., 2001: Will the North Atlantic Ocean thermohaline circulation weaken during the 21st century? Geophys. Res. Lett, 28 , 10231026.

    • Search Google Scholar
    • Export Citation
  • Goosse, H., , and T. Fichefet, 1999: Importance of ice–ocean interactions for the global ocean circulation: A model study. J. Geophys. Res, 104 , 2333723355.

    • Search Google Scholar
    • Export Citation
  • Goosse, H., , and H. Renssen, 2001: A two-phase response of the Southern Ocean to an increase in greenhouse gas concentrations. Geophys. Res. Lett, 28 , 34693473.

    • Search Google Scholar
    • Export Citation
  • Goosse, H., , E. Deleersnijder, , T. Fichefet, , and M. H. England, 1999: Sensitivity of a global coupled ocean-sea ice model to the parameterization of vertical mixing. J. Geophys. Res, 104 , 1368113695.

    • Search Google Scholar
    • Export Citation
  • Goosse, H., , F. M. Selten, , R. J. Haarsma, , and J. D. Opsteegh, 2001: Decadal variability in high Northern latitudes as simulated by an intermediate-complexity climate model. Ann. Glaciol, 33 , 525532.

    • Search Google Scholar
    • Export Citation
  • Goosse, H., , F. M. Selten, , R. J. Haarsma, , and J. D. Opsteegh, 2002: A mechanism of decadal variability of the sea-ice volume in the Northern Hemisphere. Climate Dyn, 19 , 6183.

    • Search Google Scholar
    • Export Citation
  • Goosse, H., , F. M. Selten, , R. J. Haarsma, , and J. D. Opsteegh, 2003: Large sea-ice volume anomalies simulated in a coupled climate model. Climate Dyn.,20, 523– 536.

    • Search Google Scholar
    • Export Citation
  • Gordon, A. L., 1986: Interocean exchange of thermocline water. J. Geophys. Res, 91 , 50375046.

  • Gordon, H. B., , and S. P. O'Farrell, 1997: Transient climate change in the CSIRO coupled model with dynamic sea ice. Mon. Wea. Rev, 125 , 875907.

    • Search Google Scholar
    • Export Citation
  • Gregory, J. M., , R. J. Stouffer, , S. C. B. Raper, , P. A. Stott, , and N. A. Rayner, 2002: An observationally based estimate of the climate sensitivity. J. Climate, 15 , 31173121.

    • Search Google Scholar
    • Export Citation
  • Houghton, J. T., , Y. Ding, , D. J. Griggs, , M. Noguer, , P. J. van der Linden, , X. Dai, , K. Maskell, , and C. A. Johnson, Eds.,. 2001: Climate Change 2001: The Scientific Basis. Cambridge University Press, 881 pp.

    • Search Google Scholar
    • Export Citation
  • Huybrechts, P., , and J. de Wolde, 1999: The dynamic response of the Greenland and Antarctic ice sheets to multiple-century climatic warming. J. Climate, 12 , 21692188.

    • Search Google Scholar
    • Export Citation
  • IPCC Data Distribution Center, 1999: The IPCC data distribution center CD-ROM, Version 1.0. IPCC Secretariat, Hamburg, Germany. [Available online at http://ipcc-ddc.cru.uea.ac.uk/.].

  • Joos, F., , G-K. Plattner, , T. F. Stocker, , O. Marchal, , and A. Schmittner, 1999: Global warming and marine carbon cycle feedbacks on future atmospheric CO2. Science, 284 , 464467.

    • Search Google Scholar
    • Export Citation
  • Knutti, R., , and T. F. Stocker, 2002: Limited predictability of the future thermohaline circulation close to an instability threshold. J. Climate, 15 , 179186.

    • Search Google Scholar
    • Export Citation
  • Marshall, J., , and F. Schott, 1999: Open-ocean convection: Observations, theory, and models. Rev. Geophys, 37 , 164.

  • McCarthy, J. J., , O. F. Canziani, , N. A. Leary, , D. J. Dokken, , and K. S. White, Eds.,. 2001: Climate Change 2001: Impacts, Adaptation and Vulnerability. Cambridge University Press, 1032 pp.

  • Meehl, G. A., , W. M. Washington, , D. J. Erickson, , B. P. Briegleb, , and P. J. Jaumann, 1996: Climate change from increased CO2 and direct and indirect effects of sulfate aerosols. Geophys. Res. Lett, 23 , 37553758.

    • Search Google Scholar
    • Export Citation
  • Mellor, G. L., , and T. Yamada, 1982: Development of a turbulence closure model for geophysical fluid problems. Rev. Geophys. Space Phys, 20 , 851875.

    • Search Google Scholar
    • Export Citation
  • Miller, J. R., , and G. L. Russell, 2000: Projected impact of climate change on the freshwater and salt budgets of the Arctic Ocean by a global climate model. Geophys. Res. Lett, 27 , 11831186.

    • Search Google Scholar
    • Export Citation
  • Mitchell, J. F. B., , T. C. Johns, , M. Eagles, , W. J. Ingram, , and R. A. Davis, 1999: Towards the construction of climate change scenarios. Climatic Change, 41 , 547581.

    • Search Google Scholar
    • Export Citation
  • Nakićenovic, N., and Coauthors, 2000: Special Report on Emissions Scenarios. Cambridge University Press, 599 pp.

  • Nozawa, T., , S. Emori, , A. Numaguti, , Y. Tsushima, , T. Takemura, , T. Nakajima, , A. Abe-Ouchi, , and M. Kimoto, 2001: Projections of future climate change in the 21st century simulated by the CCSR/ NIES CGCM under the IPCC SRES scenarios. Present and Future of Modeling Global Environmental Change: Toward Integrated Modeling, T. Matsuno and H. Kida, Eds., Terra Scientific, 15–28.

    • Search Google Scholar
    • Export Citation
  • Opsteegh, J. D., , R. J. Haarsma, , F. M. Selten, , and A. Kattenberg, 1998: ECBILT: A dynamic alternative to mixed boundary conditions in ocean models. Tellus, 50A , 348367.

    • Search Google Scholar
    • Export Citation
  • Rahmstorf, S., 1995: Bifurcations of the Atlantic thermohaline circulation in response to changes in the hydrological cycle. Nature, 378 , 145149.

    • Search Google Scholar
    • Export Citation
  • Renssen, H., , H. Goosse, , T. Fichefet, , and J-M. Campin, 2001: The 8.2 kyr BP event simulated by a global atmosphere–sea-ice– ocean model. Geophys. Res. Lett, 28 , 15671570.

    • Search Google Scholar
    • Export Citation
  • Renssen, H., , H. Goosse, , and T. Fichefet, 2003: On the non-linear response of the ocean thermohaline circulation to global deforestation. Geophys. Res. Lett.,30, 1061, doi:10.1029/2002GL016155.

    • Search Google Scholar
    • Export Citation
  • Roeckner, E., , L. Bengtsson, , J. Feichter, , J. Lelieveld, , and H. Rodhe, 1998: Transient climate change simulations with a coupled atmosphere–ocean GCM including the tropospheric sulphur cycle. MPI Rep. 266, 48 pp.

    • Search Google Scholar
    • Export Citation
  • Roemmich, D., , and C. Wunsch, 1985: Two transatlantic sections: Meridional circulation and heat flux in the subtropical North Atlantic Ocean. Deep-Sea Res, 32 , 619664.

    • Search Google Scholar
    • Export Citation
  • Russell, G. L., , and D. Rind, 1999: Response to CO2 transient increase in the GISS coupled model: Regional coolings in a warming climate. J. Climate, 12 , 531539.

    • Search Google Scholar
    • Export Citation
  • Schaeffer, M., , F. Selten, , and R. van Dorland, 1998: Linking IMAGE and ECBilt. RIVM Rep. 481508008, 53 pp.

  • Schaeffer, M., , F. Selten, , J. D. Opsteegh, , and H. Goosse, 2002: Intrinsic limits to predictability of abrupt regional climate change in IPCC SRES scenarios. Geophys. Res. Lett.,29, 1767, doi:10.1029/ 2002GL015254.

    • Search Google Scholar
    • Export Citation
  • Schmittner, A., , M. Yoshimori, , and A. J. Weaver, 2002: Instability of glacial climate in a model of the ocean–atmosphere–cryosphere system. Science, 295 , 14891493.

    • Search Google Scholar
    • Export Citation
  • Schmitz Jr., W. J., 1995: On the interbasin-scale thermohaline circulation. Rev. Geophys, 33 , 151173.

  • Schott, F., , M. Visbeck, , and J. Fischer, 1993: Observations of vertical currents and convection in the central Greenland Sea during winter 1988/1989. J. Geophys. Res, 98 , 1440114421.

    • Search Google Scholar
    • Export Citation
  • Smethie, W. M. J., , H. G. Ostlund, , and H. H. Loosli, 1986: Ventilation of the deep Greenland and Norwegian seas: Evidence from krypton-85, tritium, carbon-14 and argon-39. Deep-Sea Res, 33 , 675703.

    • Search Google Scholar
    • Export Citation
  • Tartinville, B., , J. M. Campin, , T. Fichefet, , and H. Goosse, 2001: Realistic representation of the surface freshwater flux in an ice-ocean general circulation model. Ocean Modell, 3 , 95108.

    • Search Google Scholar
    • Export Citation
  • van der Wal, R. S. W., , and J. Oerlemans, 1994: An energy balance model for the Greenland Ice Sheet. Global Planet. Change, 9 , 115131.

    • Search Google Scholar
    • Export Citation
  • Visbeck, M., , J. Fischer, , and F. Schott, 1995: Preconditioning the Greenland Sea for deep convection: Ice formation and ice drift. J. Geophys. Res, 100 , 1848918502.

    • Search Google Scholar
    • Export Citation
  • Washington, W. M., and Coauthors, 2000: Parallel climate model (PCM) control and transient simulations. Climate Dyn, 16 , 755774.

  • Wood, R. A., , A. B. Keen, , J. F. B. Mitchell, , and J. M. Gregory, 1999: Changing spatial structure of the thermohaline circulation in response to atmospheric CO2 forcing in a climate model. Nature, 399 , 572575.

    • Search Google Scholar
    • Export Citation
Fig. 1.
Fig. 1.

Dec–Jan–Feb (DJF) mean surface-air temperature averaged over the extratropical Northern Hemisphere (top lines, left axis) and over the GIN and Barents Seas (lower lines, right axis). A 21-yr running mean is applied to GIN and Barents Seas time series. The control simulation is indicated by a dotted line

Citation: Journal of Climate 17, 22; 10.1175/3174.1

Fig. 2.
Fig. 2.

Vertical mixing depth (solid line, left axis) and surface-air temperature (dotted line, right axis) in Jan at 77°N, 12°E for one ensemble member in scenario A1b. A 3-yr running mean is used to filter out the strongest interannual variations, while still retaining the relevant time scales

Citation: Journal of Climate 17, 22; 10.1175/3174.1

Fig. 3.
Fig. 3.

Change in annual-mean sea ice thickness (m; shaded) for 2071–2100 compared to 1961–90 for the ensemble mean. Contour lines indicate the change in convective mixing depth [contours (m) at … , −200, −100, +50]. The “extra” contour line indicates the division between areas where sea ice thickness decrease and increase

Citation: Journal of Climate 17, 22; 10.1175/3174.1

Fig. 4.
Fig. 4.

DJF mean surface-air temperature change (°C) for 2071– 2100 compared to 1961–90 for the standard experiment ensemble mean

Citation: Journal of Climate 17, 22; 10.1175/3174.1

Fig. 5.
Fig. 5.

Annual-mean surface-air temperature averaged over the GIN and Barents Seas for the relaxation experiment. The thick line indicates ensemble mean. For comparison, the ensemble mean of the standard experiment is also indicated (lower thick line)

Citation: Journal of Climate 17, 22; 10.1175/3174.1

Fig. 6.
Fig. 6.

Three-year running mean of depth of convection (lower lines, left axis) at the location of maximum convection depth in the present-day model climatology. The upper lines show sea surface salinity in the grid point just outside and north of the area of present-day convection. Solid lines represent the standard experiment and dotted lines the Arctic sea surface salinity relaxation experiment

Citation: Journal of Climate 17, 22; 10.1175/3174.1

Fig. 7.
Fig. 7.

Salinity for the ocean layer at 1500–3500 m for four consecutive 30-yr mean periods for one ensemble member in the standard experiment. The numbers in the 2071–2100 panel refer to the location of the points in the time series plot of Fig. 8

Citation: Journal of Climate 17, 22; 10.1175/3174.1

Fig. 8.
Fig. 8.

Salinity in seven selected ocean grid cells in the layer at 1500–3500 m (left axis). The geographical location of the grid cells is indicated in the bottom-right panel of Fig. 7. Also shown in mean surface salinity in the Arctic Ocean (“arctic,” right axis)

Citation: Journal of Climate 17, 22; 10.1175/3174.1

Fig. 9.
Fig. 9.

Semiequilibrium climate states characterized by long-term mean patterns of deep convection. Depth of convection (contour distance 100 m) is shown for 30-yr mean periods from (top) the 50% control run (state I) and from 250 yr after the freshwater pulse in (middle) the 30% experiment (state II) and (bottom) 10% experiment (state III)

Citation: Journal of Climate 17, 22; 10.1175/3174.1

Fig. 10.
Fig. 10.

Annual mean GIN Sea deep-water formation for the standard 50% experiment and the 30% and 10% experiments. For convenience, the start of the historical spinups is labeled 1851, so that the freshwater pulse in the 30% experiment is labeled 1651

Citation: Journal of Climate 17, 22; 10.1175/3174.1

Fig. 11.
Fig. 11.

Dec–Jan–Feb mean surface-air temperature change (°C) for 2071–2100 compared to 1961–90 for the 30% ensemble mean

Citation: Journal of Climate 17, 22; 10.1175/3174.1

Fig. 12.
Fig. 12.

Change in annual mean sea ice thickness (shaded) for 2071–2100 compared to 1961–90 for the 30% ensemble mean. Contour lines indicate the change in convective mixing depth, with a contour distance of 100 m. The “extra” contour line indicates the division between areas where sea ice thickness decrease and increase

Citation: Journal of Climate 17, 22; 10.1175/3174.1

Save