• Appenzeller, C., , T. F. Stocker, , and M. Anklin, 1998: North Atlantic oscillation dynamics recorded in Greenland ice cores. Science, 282 , 446449.

    • Search Google Scholar
    • Export Citation
  • Basnett, T. A., , and D. E. Parker, 1997: Development of the Global Mean Sea Level Pressure Data Set GMSLP2. Hadley Centre Climate Research Tech. Note 79, Met Office, Bracknell, Berkshire, United Kingdom.

    • Search Google Scholar
    • Export Citation
  • Cattle, H., , and J. Crossley, 1995: Modelling Arctic climate change. Philos. Trans. Roy. Soc. London, A352 , 201213.

  • Collins, M., , S. F. B. Tett, , and C. Cooper, 2001: The internal climate variability of HadCM3, a version of the Hadley Centre coupled model without flux adjustments. Climate Dyn, 17 , 6181.

    • Search Google Scholar
    • Export Citation
  • Cook, E. R., , R. D. D'Arrigo, , and K. R. Briffa, 1998: A reconstruction of the North Atlantic oscillation using tree-ring chronologies from North America and Europe. Holocene, 8 , 917.

    • Search Google Scholar
    • Export Citation
  • Cox, M. D., 1984: A primitive equation, 3 dimensional model of the ocean. GFDL Ocean Group Tech. Rep. No. 1, Princeton, NJ, 143 pp.

  • Curry, R., , and M. S. McCartney, 2001: Ocean gyre circulation changes associated with the North Atlantic oscillation. J. Phys. Oceanogr, 31 , 33743400.

    • Search Google Scholar
    • Export Citation
  • Curry, R., , M. S. McCartney, , and T. M. Joyce, 1998: Oceanic transport of subpolar climate signals to mid depth subtropical waters. Nature, 391 , 575577.

    • Search Google Scholar
    • Export Citation
  • Czaja, A., , and C. Frankignoul, 1999: Influence of the North Atlantic SST on the atmospheric circulation. Geophys. Res. Lett, 26 , 29692972.

    • Search Google Scholar
    • Export Citation
  • Deser, C., , and M. L. Blackmon, 1993: Surface climate variations over the North Atlantic Ocean during winter: 1900–1989. J. Climate, 6 , 17431753.

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

    • Search Google Scholar
    • Export Citation
  • Gordon, C., , C. Cooper, , C. A. Senior, , H. T. Banks, , J. M. Gregory, , T. C. Johns, , J. F. B. Mitchell, , and R. A. Wood, 2000: The simulation of SST, sea ice extents and ocean heat transports in a version of the Hadley Centre coupled model without flux adjustments. Climate Dyn, 16 , 147168.

    • Search Google Scholar
    • Export Citation
  • Hall, M. M., , and H. L. Bryden, 1982: Direct estimates and mechanisms of ocean heat transport. Deep-Sea Res, 29 , 339359.

  • Hurrell, J. W., 1995: Decadal trends in the North Atlantic oscillation: Regional temperatures and precipitation. Science, 269 , 676679.

    • Search Google Scholar
    • Export Citation
  • Latif, M., 1998: Dynamics of interdecadal variability in coupled ocean–atmosphere models. J. Climate, 11 , 602624.

  • Levitus, S., , and T. P. Boyer, 1994: Temperature. Vol. 4, World Ocean Atlas 1994, NOAA Atlas NESDIS 4, 117 pp.

  • Levitus, S., , R. Burgett, , and T. P. Boyer, 1994: Salinity. Vol. 3, World Ocean Atlas 1994, NOAA Atlas NESDIS 3, 99 pp.

  • Macdonald, A. M., , and C. Wunsch, 1996: Oceanic estimates of global ocean heat transport. U.S. WOCE Rep. 8, 28–30.

  • Parker, D. E., , C. K. Folland, , and M. Jackson, 1995: Marine surface temperature: observed variations and data requirements. Climatic Change, 31 , 559600.

    • Search Google Scholar
    • Export Citation
  • Pope, V. D., , M. Gallani, , P. R. Rowntree, , and R. A. Stratton, 2000: The impact of new physical parameterisations in the Hadley Centre climate model—HadAM3. Climate Dyn, 16 , 123146.

    • Search Google Scholar
    • Export Citation
  • Rayner, N. A., , E. B. Horton, , D. E. Parker, , C. K. Folland, , and R. B. Hackett, 1996: Global Sea-Ice and Sea Surface Temperature data set, 1903–1994. Version 2.2, Hadley Centre for Climate Prediction and Research Tech. Note 74, Met Office, Bracknell, Berkshire, United Kingdom.

    • Search Google Scholar
    • Export Citation
  • Roberts, M. J., , R. Marsh, , A. L. New, , and R. A. Wood, 1996: An intercomparison of a Bryan–Cox-type ocean model and an isopycnic ocean model. Part I: The subpolar gyre and high-latitude processes. J. Phys. Oceanogr, 26 , 14951527.

    • Search Google Scholar
    • Export Citation
  • Rodwell, M. J., , and C. K. Folland, 2001: Atlantic air–sea interaction and seasonal predictability. Quart. J. Roy. Meteor. Soc., in press.

    • Search Google Scholar
    • Export Citation
  • Rodwell, M. J., , D. P. Rowell, , and C. K. Folland, 1999: Oceanic forcing of the winter North Atlantic oscillation and European climate. Nature, 398 , 320323.

    • Search Google Scholar
    • Export Citation
  • Seager, R., , Y. Kushnir, , M. Visbeck, , N. Naik, , J. Miller, , G. Krahmann, , and H. Cullen, 2000: Causes of Atlantic Ocean climate variability between 1958 and 1998. J. Climate, 13 , 28452862.

    • Search Google Scholar
    • Export Citation
  • Semtner, A. J., 1976: A model for the thermodynamic growth of sea ice in numerical investigations of climate. J. Phys. Oceanogr, 6 , 379389.

    • Search Google Scholar
    • Export Citation
  • Sutton, R. T., , and M. R. Allen, 1997: Decadal predictability of North Atlantic sea surface temperatures and climate. Nature, 388 , 563567.

    • Search Google Scholar
    • Export Citation
  • Wood, R. A., , A. B. Keen, , J. F. B. Mitchell, , and J. M. Gregory, 1999:: Localised collapse of the thermohaline circulation in a climate model with increasing atmospheric CO2. Nature, 399 , 572575.

    • Search Google Scholar
    • Export Citation
  • Wright, D. K., , and C. Gordon, 1997: The positioning of the North Atlantic Current. Ocean Applications Tech. Note 14, Met Office, Bracknell, Berkshire, United Kingdom.

    • Search Google Scholar
    • Export Citation
  • View in gallery

    (a) Observed NAO index based on normalized wintertime (DJF) sea level pressure differences between Lisbon (Portugal) and Stykkisholmur (Iceland), defined and smoothed as in Hurrell (1995). Heavy line is the smoothed index. (b) Composite observed wintertime mean sea level pressure pattern, calculated as the mean of years NAOI > 1 minus the mean of years when NAOI < 1. Both figures calculated using the GMSLP (Basnett and Parker 1997) dataset

  • View in gallery

    North Atlantic sea surface temperature (°C): (a) HadCM3 control and (b) GISST 2.2 observed annual mean climatology. (c) HadCM3 50-m currents (cm s−1) (only currents above 4 cm s−1 are plotted), (d) 670-m currents (only currents above 1 cm s−1 are plotted), and (e) 2100-m currents (only currents above 1 cm s−1 are plotted). Model data based on annual mean of years 540–639

  • View in gallery

    (Continued)

  • View in gallery

    (Continued)

  • View in gallery

    Standard deviation of SST (°C) from (a) MOHSST (Parker et al. 1995) observed dataset, 1948–94, and (b) the HadCM3 control, years 540–639. Note the model data have been masked after calculating standard deviations for ease of comparison with the observational data

  • View in gallery

    Observed wintertime (DJF) sea surface temperature anomalies (°C) along the climatological path of the NAC, as in Sutton and Allen (1997) Fig. 2a, but calculated using GISST version 3.0 dataset. No smoothing applied. Contour interval is 0.4°C and differences greater than 0.2°C are shaded

  • View in gallery

    NAO index for the first 1000 yr of the HadCM3 control experiment, calculated using normalized wintertime (DJF) sea level pressure differences between Lisbon (Portugal) and Stykkisholmur (Iceland), defined and smoothed as in Hurrell (1995). Heavy line is the smoothed index

  • View in gallery

    (a) NAO index, as in Fig. 5, for year 540–639 of the HadCM3 control run. (b) Composite wintertime mean sea level pressure for year 540–639 of the HadCM3 control, calculated as the mean of years NAOI > 1 minus the mean of years when NAOI < 1

  • View in gallery

    Mean temperature (°C), solid line, and salinity (psu), dotted line, in the Labrador Sea between 1000 and 1500 m. Annual mean data, years 540–639 of the HadCM3 control simulation. The smoothed NAOI from Fig. 6 is also shown for ease of comparisons as the dashed line

  • View in gallery

    Model wintertime (DJF) sea surface temperature (°C) anomalies, years 540–593, (relative to 540–639 mean) along the mean path of the model NAC. The “X” marks the region in which the subpolar and subtropical gyres meet (off Newfoundland). No smoothing has been applied. Contour interval is 1.0°C and differences greater than 0.5°C are shaded

  • View in gallery

    Unsmoothed NAO index for control (solid line) as in Fig. 5. CONV (dashed line) and INHIB (dotted line)

  • View in gallery

    Sea surface temperature difference (°C) for (a) CONV minus HadCM3 and (b) INHIB minus HadCM3. Contour interval is 1.0°C and differences greater than 1.0°C are shaded. Sea surface salinity difference (psu) for (c) CONV minus HadCM3, and (d) INHIB minus HadCM3. Contour interval is 0.5 psu and differences greater than 0.5 psu are shaded. Wintertime mean (DJF) of years 10–14 of sensitivity studies, equivalent years for the HadCM3 control. The region over which salt forcing is applied is shown in (a) as a rectangular box

  • View in gallery

    1000-m temperature difference (°C) for (a) CONV minus HadCM3 and (b) INHIB minus HadCM3. Contour interval is 0.4°C and differences greater than 0.2°C are shaded. 1000-m salinity difference (psu) for (c) CONV minus HadCM3 and (d) INHIB minus HadCM3. Contour interval is 0.04 psu and differences greater than 0.02 psu are shaded. Wintertime mean (DJF) of years 10–14 of sensitivity studies, equivalent years for the HadCM3 control

  • View in gallery

    50-m currents (cm s−1) for (a) CONV and (b) INHIB. Both are wintertime (DJF) averaged over years 10–14. Only current strengths above 4 cm s−1 are plotted

  • View in gallery

    1000-m currents (cm s−1) for (a) CONV and (b) INHIB. Both are wintertime (DJF) averaged over years 10–14. Only current strengths above 1.5 cm s−1 are plotted

  • View in gallery

    Potential density (σ0) cross section at 50°N. Wintertime mean, year-10 (a) CONV, (b) INHIB, and year-10 equivalent (c) HadCM3 control

  • View in gallery

    (a) Thickness of isopycnal layer (27.9 < σ0 < 28.0) at 50°N from CONV. Wintertime means (DJF), (b) latitudinal standard deviation (m) of this layer thickness at 50°N in HadCM3, (c) SST anomalies (°C) at 50°N (relative to 50-yr wintertime mean of HadCM3, temperature anomalies above 2°C are shaded), and (d) latitudinal standard deviation of SST anomaly at 50°N from HadCM3 (°C). All standard deviations are calculated for wintertime means over the same 50-yr period used in calculating the anomalies

  • View in gallery

    (Continued)

  • View in gallery

    (a) Wintertime mean SST difference (°C) between CONV and HadCM3 plotted along the pathway of the mean HadCM3 NAC. The “X” marks the region in which the subpolar and subtropical gyres meet (off Newfoundland). Contour interval is 1.0°C, negative contours are dotted, and differences greater than 2.0°C are shaded. (b) Standard deviation of wintertime SST along the pathway (°C), calculated using the same 50-yr period using to derive anomalies

  • View in gallery

    Wintertime (DJF) mean currents at 300 m (cm s−1) for (a) CONV and (b) INHIB, year 7, and for (c) CONV and (d) INHIB, year 12. Only current strengths above 4.0 cm s−1 are plotted

  • View in gallery

    Modeled potential temperature (°C) of LSW, averaged between 1000 and 1500 m (solid line), in the cyclic forcing experiment. The salinity that was relaxed toward throughout the experiment, in the forcing region, is indicated by the dotted line. Vertical dashed lines mark the start of each new cycle of forcing

  • View in gallery

    Power spectra of SST anomaly on a section of the NAC (43°N, 41°–47°W). The solid line is the control experiment, dashed line the cyclic forcing experiment

  • View in gallery

    Cyclic forcing experiment, detrended and filtered (retaining 7–20 yr) wintertime SST anomalies (°C) along the mean path of the modeled NAC (taken from HadCM3 control). The “X” marks the region in which the subpolar and subtropical gyres meet (off Newfoundland). Contour interval is 0.5°C, with 0.25°C contour added, and differences greater than 0.25°C are shaded

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 10 10 1
PDF Downloads 5 5 2

North Atlantic Oceanic Decadal Variability in the Hadley Centre Coupled Model

View More View Less
  • 1 Ocean Applications, Hadley Centre for Climate Prediction and Research, Met Office, Bracknell, Berkshire, United Kingdom
© Get Permissions
Full access

Abstract

The Hadley Centre coupled model is used to investigate the relationship between the North Atlantic oscillation (NAO), Labrador Sea deep convection, and sea surface temperature variability. It is shown that the model is capable of simulating many features of the observed variability. In particular, the model reproduces the observed relationship between decadal variations in Labrador Sea convection and the NAO. It also has surface temperature anomalies that propagate along the path of the North Atlantic Current (NAC) with a timescale close to that of observed anomalies. A number of sensitivity experiments are performed with the coupled model to understand the underlying mechanisms. Simulations in which the Labrador Sea convection is artificially enhanced and surpressed illustrate a time-lagged impact on the NAC and therefore on sea surface temperature in that region. Further simulations with a sinusoidal forcing of convection confirm the role of Labrador Sea Water (LSW) in determining the timescale for propagation of the surface temperature anomalies. The timescale of surface anomalies moving along the path of the NAC is determined by the movement of LSW at intermediate depths.

Corresponding author address: Claire Cooper, Ocean Applications, Hadley Centre for Climate Prediction and Research, Met Office, London Road, Bracknell, Berkshire, RG12 2SY United Kingdom. Email: claire.cooper@metoffice.com

Abstract

The Hadley Centre coupled model is used to investigate the relationship between the North Atlantic oscillation (NAO), Labrador Sea deep convection, and sea surface temperature variability. It is shown that the model is capable of simulating many features of the observed variability. In particular, the model reproduces the observed relationship between decadal variations in Labrador Sea convection and the NAO. It also has surface temperature anomalies that propagate along the path of the North Atlantic Current (NAC) with a timescale close to that of observed anomalies. A number of sensitivity experiments are performed with the coupled model to understand the underlying mechanisms. Simulations in which the Labrador Sea convection is artificially enhanced and surpressed illustrate a time-lagged impact on the NAC and therefore on sea surface temperature in that region. Further simulations with a sinusoidal forcing of convection confirm the role of Labrador Sea Water (LSW) in determining the timescale for propagation of the surface temperature anomalies. The timescale of surface anomalies moving along the path of the NAC is determined by the movement of LSW at intermediate depths.

Corresponding author address: Claire Cooper, Ocean Applications, Hadley Centre for Climate Prediction and Research, Met Office, London Road, Bracknell, Berkshire, RG12 2SY United Kingdom. Email: claire.cooper@metoffice.com

1. Introduction

Recently there has been considerable progress in the understanding of observed ocean climate variability in the North Atlantic on decadal timescales. Many of the subsurface ocean changes over the last 50 years appear to be driven by large decadal swings in the North Atlantic oscillation (NAO; Dickson et al. 1996; Curry et al. 1998). A picture is emerging of the basin-scale North Atlantic Ocean variability being forced by wind and buoyancy forcing associated with the NAO. This can explain much of the observed decadal and interdecadal changes in deep convection in the Labrador and Greenland Seas (Dickson et al. 1996), variations in the strength of the North Atlantic Current (NAC; Curry and McCartney 2001) and, in large part, the tripole pattern of sea surface temperature (SST) variability (Deser and Blackmon 1993; Seager et al. 2000).

The time series of the NAO index [NAOI; calculated as the normalized winter pressure difference between Stykkisholmur (Iceland) and Lisbon (Portugal)] and the composite pattern of pressure at mean sea level (PMSL) variability over the North Atlantic basin (calculated as the mean of years NAOI > 1 minus the mean of years NAOI < 1) from observations [Global Mean Sea Level Pressure dataset (GMSLP; Basnett and Parker 1997)] is shown in Fig. 1. Figure 1a shows the large change in NAO index over the last 30 years, the cause of which has yet to be established. It may be due to natural variability, a consequence of anthropogenic climate change, or both. It does appear from modeling studies (Rodwell et al. 1999) that to some degree the NAO response is being forced by North Atlantic SST variability.

Recent analyses of data by Czaja and Frankignoul (1999) and Rodwell and Folland (2001) suggest that the SST anomalies responsible for (weakly) forcing the North Atlantic atmosphere do not necessarily correspond to the tripole pattern. Czaja and Frankignoul (1999) use Comprehensive Ocean–Atmosphere Data Set (COADS) SST data and the National Center for Atmospheric Research (NCAR) 500-mb heights in a singular vector decomposition (SVD) analysis to identify coupled modes that are lagged in time. For the atmosphere leading the ocean by one month, or less, the dominant SST pattern is the tripole. For the ocean leading the winter atmosphere by three months they find that it is an SST pattern similar to the tripole that forces an atmospheric response over the northwestern Labrador Sea in late spring. The NAO in early winter is found to be forced by the preceding summer SSTs, with centers of action in the SST pattern located southeast of Newfoundland and in the eastern subtropics. This pattern is seasonal, it is not related to the SST tripole, and its centers of action do not coincide with those of the tripole. Rodwell and Folland (2001) performed a similar analysis using different data sources and found significant forcing of the atmosphere at certain times of year. These studies suggest that there is some seasonal timescale atmospheric predictability from SST anomalies in the North Atlantic.

Any possible longer timescale predictability will require prediction of the SSTs in the regions that may force the atmosphere. If this is possible at all, the long timescales are likely to arise because of the role of ocean dynamics rather than simply local air–sea interactions. Seager et al. (2000) have shown that the tripole SST pattern can largely be explained without evoking ocean dynamics (other than local Ekman advection). However, as already noted, this is not necessarily the most important pattern in forcing the atmosphere.

This paper investigates the existence and causal mechanisms of decadal SST anomalies in the NAC region as simulated in the latest Hadley Centre coupled climate model (HadCM3). In order to do this it is first necessary to determine the degree of realism of decadal timescale variability in the coupled model simulation. In particular, the increasing number of observed relationships between the NAO and ocean conditions discovered in recent years need to be studied in the model simulation. Only if the model is capable of realistically modeling these features can we have some confidence in the model's ability to represent underlying mechanisms. It is the detailed modeling and understanding of these mechanisms that may, in the future, enable some degree of predictability of natural climate variability.

This paper is organized as follows. In section 2 the coupled model is described and its climatology for the North Atlantic ocean is shown in some detail. Section 3 summarizes some of the recent observational results relating ocean variability to the NAO, and section 4 considers the same relationships in the coupled model simulation. To better elucidate the mechanisms of SST variability associated with the NAC, section 5 describes some sensitivity experiments carried out with the coupled model. Section 6 contains a discussion of the results and is followed by some conclusions in section 7.

2. Model description and climatology

The model, HadCM3, is a coupled atmosphere–ocean–sea ice general circulation model developed at the Hadley Centre for use in climate studies (Gordon et al. 2000). The atmospheric model has a horizontal grid spacing of 3.75° × 2.5° and 19 vertical levels with detailed parameterizations of physical processes. A full description of the atmospheric model can be found in Pope et al. (2000).

The ocean component is a 20-level version of the Cox (1984) model with a horizontal grid spacing of 1.25° × 1.25°. There are six ocean grid boxes to every atmospheric grid box and the land–sea masks match exactly. Vertical levels are distributed to enhance resolution near the ocean surface. There are detailed parameterizations of along isopycnal and vertical mixing. The sea-ice model is based on Semtner's (1976) zero-layer treatment of ice thermodynamics with a parameterization of leads and simple ice advection by ocean surface currents. The sea ice model runs at the same resolution as the ocean model. Further details of the ocean and sea-ice models can be found in Gordon et al. (2000) and Cattle and Crossley (1995).

Ocean currents are initially set to zero, with initial temperature and salinity fields taken from the Levitus and Boyer (1994) climatology. The atmospheric initial state is taken from an earlier atmosphere-only model integration. The ocean–atmosphere coupling period is one day. The HadCM3 control experiment uses preindustrial atmospheric trace gas concentrations and has been run without the use of flux adjustments for over 1000 years without appreciable drift in the model's climate (less than 0.009°C century−1 change in global mean SST).

After five centuries of coupled simulation, the model representation of North Atlantic SSTs is shown in Fig. 2a. For comparison Global Ice and Sea Surface Temperatures (GISST), version 2.2, (Rayner et al. 1996) observed climatological SSTs are shown in Fig. 2b. It can be seen that the SST simulation is generally comparable with climatology although the region of high SST gradient to the south of Newfoundland that is associated with the NAC (see Figs. 2c and 2d) is too far north in the model and the current turns northward in the wrong place. These errors are associated with the fact that the model Gulf Stream does not separate from the North American coast at Cape Hatteras, as it does in reality, but leaves the coast farther north. The North Atlantic Current also tends to be too zonal. Both of these problems are common among models of this resolution (Roberts et al. 1996). Figures 2c and 2d also show the simulation of the subpolar gyre and demonstrate that the Labrador and Greenland Currents are represented. Figure 2e shows the deep currents at 2100 m and, in particular, the Deep Western Boundary Current (DWBC) showing the model flow at this depth out of the subpolar region. Taken together, the modeled fields in Fig. 2 illustrate that many of the gross features of the North Atlantic Ocean circulation are reproduced in the model. Although not shown here, there is a long-term drift in the model (both in temperature and salinity) within the Atlantic basin. After 600 years the temperatures around 1000 m are 2.5°C warmer than those of Levitus and Boyer (1994) and the salinity is 0.6 ppt too salty. These errors remain fairly constant throughout the last 400 years of the simulation. The reasons for these drifts have yet to be fully understood. A consequence of these errors is that mean watermass properties, such as that of the modeled Labrador Sea Water (LSW), are too warm and salty. There is, however, a compensation in density between the errors resulting from this drift.

A major improvement of the HadCM3 model over earlier versions, which employed a coarser resolution ocean component, is the more realistic simulation of the geographical regions of deep convection in the Labrador and Greenland Seas (Wood et al. 1999). The maximum strength of the North Atlantic meridional overturning circulation occurs at approximately 45°N and has an average value of 19 Sverdrups, which compares well with observed estimates (Hall and Bryden 1982). The modeled mean poleward heat transport across 24°N in the Atlantic is 1.0 PW (1015 W) as compared with the observational estimates of 1.2 (Hall and Bryden 1982) and 1.1 PW (Macdonald and Wunsch 1996). Further details of the control integration and its climatology can also be found in Gordon et al. (2000).

The standard deviations of SST calculated from the Met Office Historical Sea Surface Temperature (MOHSST) dataset gridded in 1° boxes (Parker et al. 1995) and the model are shown in Fig. 3. MOHSST is used here rather than GISST as the latter has been created using an empirical orthogonal function (EOF) reconstruction technique to fill in missing data points, which has a consequence of smoothing the fields in the high SST gradient region associated with the NAC. The modeled standard deviations compare well in magnitude with the observations, with both model and data showing highest variability along the Gulf Stream/NAC system. This is to be expected because this is a region of very high horizontal SST gradient and small shifts in the position of the NAC will lead to significant surface temperature anomalies. The model overestimates the spatial extent of the high variability, particularly over the subpolar gyre region, when compared with the available observations.

3. Ocean response to NAO—Observations

To a large extent the forcing functions of momentum, heat, and freshwater at the surface in the North Atlantic will reflect the variability in the NAO. There has been a considerable amount of information collected concerning observed changes in the North Atlantic Ocean over the last few decades. Dickson et al. (1996) have shown that many of these data can be combined to illustrate that coordinated changes have been occurring in the region over this period. They demonstrated that there have been major changes in the water masses produced in the deep convection regions of both the Labrador and Greenland Seas over the last 50 years and that these changes can be linked with variations in NAO forcing. These two high-latitude seas are important, especially for climate, because of their role as a source for the deep water that contributes to driving the global thermohaline circulation.

Dickson et al. (1996) studied variability in the observational record of convective sites in the North Atlantic. They collated annual average temperature and salinity data between 1000 and 1500 dbar from OWS BRAVO, the core of the LSW layer, and noted that there has been a marked decrease in both quantities since the 1970s. A cooling in this region, and at this depth, is indicative of increased deep convection. They also noted that the NAO and LSW temperature time series are anticorrelated over the last 40 years, that is, that deep convection in the Labrador Sea is positively correlated with the NAO index. In fact, Labrador Sea convection effectively switched off in the late 1960s/early 1970s, when the NAO was at a local minimum. Since then it has increased to reach record strength during the 1990s, which coincides with the NAO index reaching a local maximum. Dickson et al. (1996) also showed that there is an anticorrelation over the same period between deep convection in the Greenland Sea and the NAO. Further, when deep convection was at a minimum in the Labrador Sea it was near a maximum in the Greenland Sea.

Curry et al. (1998) have taken this a step further and considered the implications of changes in Labrador Sea deep convection on the circulation of the North Atlantic Ocean. They have shown that LSW can be traced from its source in the subpolar gyre to two destinations. Some of the water that sinks to deep levels in the Labrador Sea travels south, in the DWBC beneath the Gulf Stream. It takes approximately 7 yr for the water to move from the Labrador Sea to deep levels in the subtropics east of Florida. In another branch, the LSW gets caught in the NAC and is advected northeastward at midlevels in the ocean. The consequences of variability in LSW and deep convection, as described above, are clearly seen in both the western subtropics at depth and in the central and eastern North Atlantic, although lagged in time.

Sutton and Allen (1997) analyzed observed wintertime North Atlantic SSTs and calculated lagged correlations over the entire basin with a source region in the vicinity of Cape Hatteras, where the Gulf Stream separates from the North American coast. They demonstrated that at least one region could be found that showed a correlation higher then 0.8 with the source region for lags of up to 7 yr. Beyond this point the correlations dropped slightly. The region of high correlation moved across the basin from west to east as the lag increased. This indicated that an anomaly originating near Florida in the subtropics would appear to propagate along the path of the NAC, taking approximately 9 yr to cross the basin. This is much slower than the advective speed of the NAC. To demonstrate this they defined a climatological path for the Gulf Stream and NAC and then plotted observed SST anomalies (1945–89) along the pathway (Fig. 2a of Sutton and Allen 1997). Figure 4 shows an updated version of their figure. The pathway along which anomalies are plotted is the same as that used by Sutton and Allen (1997) while the observed temperatures are taken from the GISST 3.0 dataset, an updated version of GISST2.2 (Rayner et al. 1996). No smoothing has been applied in this case (although there is some spatial smoothing implicit in the construction of the GISST dataset). We use GISST rather than MOHSST in order to take advantage of the global data coverage of the dataset. Warm and cold anomalies are clearly seen propagating along the path of the NAC. There is a very clear warm anomaly that dominates the period around 1950. This is followed by a cold anomaly appearing in the 1960s. These coincide with times when the NAO index was at local maximum and minimum, respectively, although with such a relatively short data record it is not possible to deduce whether this can be attributed to anything other than coincidence.

In this paper we use a coupled model to explore the relationship between NAO, deep convection, NAC, and SST variability. The observational data is relatively sparse at depth so a coupled model is a very useful tool in attempting to understand relationships between these quantities. It should also be noted that all of the original work done in the observational studies described above are based on only 50 years of data whereas the model can provide much longer data records. The short observational subsurface data record means that few of the above relationships could be regarded as statistically significant. Experiments with coupled models can help to establish the robustness of these relationships.

4. Model simulation compared with observations

Over the last 10 years a considerable amount of work has been carried out on the analysis of decadal variability in coupled models [see Latif (1998) for a review], although very little has been done to compare the models with observations other than at the surface. Despite the sparse observations it can be seen from the previous section that there is a reasonably detailed description of some facets of North Atlantic decadal variability emerging from the data. In this section we will compare key aspects of variability in HadCM3 with observations.

a. The North Atlantic oscillation

The control integration of HadCM3 has completed over 1000 years and Fig. 5 shows the NAOI from the coupled model for the first 1000 years of the simulation. The index is defined in the same way as Fig. 1a following Hurrell (1995). Note that while there are periods of relatively little decadal variability in the index there are also periods (e.g., from years 650–670) that show changes of a similar magnitude to those observed in the NAOI over the last few decades. This indicates, in the model at least, that these large changes are possible without invoking the need for anthropogenic effects. It should be noted however that analysis of the model NAOI by Collins et al. (2001) has shown that, while the magnitude of the observed changes in the NAO are within the variability of HadCM3 control run, the rapidity of change is not. They conclude that the observed rate of increase is highly unusual in comparison with the 1000 years of HadCM3 NAOI. Collins et al. (2001) also found that the power spectra of the model and the observations are both consistent with a white noise process and have no significant spectral peaks. The model and observational spectra are also found to be statistically consistent.

The considerable century timescale variability in the decadal variations of the modeled NAO is not unlike the behavior seen in both the observation record and in proxies for the NAOI in paleo-based reconstructions (Appenzeller et al. 1998; Cook et al. 1998). These proxies show that the recently observed large shift between phases of the NAO was preceded by a period of relatively little decadal variability, that is, 1870–1900.

Because our intention is to compare the model with observations from the last 50 years, it is appropriate to chose a period of the control in which NAO variability is similar to that in the ocean observed record, that is, a period of sustained negative NAO forcing followed by sustained positive NAO forcing. For this reason a 100-yr period starting at model year 540 was chosen. This is a time of high decadal variability in the model with variations similar in both magnitude and timescale to the NAO variability that has been observed over recent decades (see Fig. 1a). The time series of NAOI for this period is shown in Fig. 6a. The spatial pattern is illustrated by the PMSL composite (NAO+ minus NAO−) in Fig. 6b. This can be compared with the equivalent pattern from observations shown earlier in Fig. 1b. The general north–south dipole is well reproduced by the model although the southern center of action is located more over the central Atlantic than over Portugal and Spain.

In what follows we consider two aspects of the ocean response in the HadCM3 control simulation during this period, first, the Labrador Sea deep convection and then anomalies propagating along the NAC.

b. Labrador Sea deep convection

As already mentioned, one feature of HadCM3 that is an improvement over earlier model versions is a more realistic simulation of the sites of deep convection in the North Atlantic. Previous model versions, with a coarser-resolution ocean component, tended to have North Atlantic deep convection occurring in one location, a relatively small area in the Irminger Sea, south of Greenland. In HadCM3 deep convection occurs in the more realistic locations of the Labrador Sea and Greenland Sea. As described above, Dickson et al. (1996) have linked observed changes in deep convection at these sites with variations in NAO forcing. The better representation of deep convection in this version of the Hadley Centre model means that we can look explicitly at the relationship between NAO and deep convection.

Figure 7 shows the mean temperature and salinity in the Labrador Sea between 1000- and 1500-m depth for the 100-yr period used to calculate the NAOI in Fig. 6a. In fact, a smoothed NAOI is replotted in Fig. 7 for ease of comparison. In the model, as in observations, the depth-averaged temperature and salinity, that is, the properties of LSW, can be taken as indicators of changes in the rate of deep convection. Hence as temperature and salinity at this depth is decreasing, deep convection and the thickness of LSW are increasing. A clear decadal variation can be seen in Fig. 7 in both temperature and salinity. Note that due to the drift mentioned in section 2 the modeled mean watermass properties are such that the modeled LSW is too warm and saline. The NAOI and these deep convection indicators show a clear anticorrelation relationship between the smoothed time series (r = −0.46). Therefore in the model, as in the observations, there is a positive correlation between the NAO and Labrador Sea convection. It should be noted, however, that there are other periods of the control simulation (not shown), with weak NAO variability on decadal timescales when there is very little correlation between these quantities.

c. Propagating SST anomalies

The model also simulates propagating SST anomalies similar to those found in the observational data by Sutton and Allen (1997). In their paper Sutton and Allen plot SST anomalies along the climatological path of the NAC. In the model the NAC separation from the North American coast occurs too far north and the current is generally too zonal (see Fig. 2c). For this reason the mean modeled track of the NAC is used to define the along path coordinate for model data rather that the climatological path. The model SST anomalies along this path are shown in Fig. 8. The range over which the model data are plotted has been reduced from the full 100 years to be the same length as the observational record in order to keep the aspect ratio the same as in Fig. 4. However, the model SST anomalies, years 540–585, shown in Fig. 8 can be considered representative of the entire 100-yr period. This use of a smaller sample is purely to ease comparison of propagation timescales. As in Fig. 4 no smoothing has been applied to the data. Warm and cold anomalies can clearly be seen propagating through from west to east in the model data, that is, the cold anomaly that begins around model year 540 or the warm anomaly beginning around year 565. The speed of propagation of these anomalies in the observations and model data are remarkably close, both being much slower than the advective timescale for crossing the basin. (The banded structure highlighted near the center of Fig. 8 by a line marked “X” is in the region of the pathway in which waters from the subtropical and subpolar gyres meet.)

5. Labrador Sea convection experiments

To investigate the possible links between Labrador Sea convection and SST anomalies in the subpolar gyre a number of sensitivity studies have been performed with the coupled model, HadCM3. Because we are principally interested in SST responses it is desirable to use a coupled model, with its comprehensive representation of air–sea interaction processes, rather than a forced ocean-only model. For the validation study described in the previous section it was desirable to study a period of the control integration that exhibited strong decadal variability. This meant that any signals in the ocean would be as large as possible, as well as being comparable with the NAO forcing observed over the last 50 years. For sensitivity studies on the effects of convection in the Labrador Sea, a period of the HadCM3 control integration with very little decadal variability in the NAOI was taken as a start point in order to reduce the level of background decadal variability. All of the experiments described in what follows begin from year 400 of the control and run parallel to it from this point. It can be seen in Fig. 5 that this is a period of relatively low NAO activity in the HadCM3 control.

a. Continuous forcing of Labrador Sea convection

Two experiments were designed in order to assess the effects of enhanced or inhibited convection. In order to force such events we utilized additional strong relaxation on salinity, with a restoring timescale of approximately 15 days for a 50-m-layer, applied only to the small region of the Labrador Sea that is associated with deep convection in the model (58.125°–60.625°N, 58.75°–50°W, as shown in Fig. 10a). Controlling the deep convection by forcing the salinity field means that, within HadCM3, there is no mechanism for a direct local thermal feedback to the atmosphere. Addition of an anomalous heat flux would hold the potential to directly change atmospheric circulation through local feedbacks and may thereby confuse the response.

In the first experiment, denoted CONV, enhanced salinity forcing was applied continually to the forcing region of the Labrador Sea in order to promote deep convection. A background value of 35.4 psu was set for the relaxation. In the second sensitivity experiment, hereinafter INHIB, a freshwater (reduced salinity) forcing was applied to have a stabilizing effect on the water column and hence inhibit deep convection, a background value of 32 psu being set. The minimum and maximum values of surface salinity in the Levitus et al. (1994) climatology for this region are 33.0 and 34.8 psu, respectively (based on monthly mean data). In the HadCM3 control period used here, there is a range of 31.4–35.3 psu in wintertime means of surface salinity, the wintertime mean value for the entire period being 33.7 psu. Thus, while the range of forcing in CONV and INHIB (which is centered about the mean control value) is beyond the range of climatological values, the forcing chosen is not unrealistic for the coupled model. The experiments were run for 14 years. Figure 9 shows the evolution of the NAO index in the control and the two sensitivity experiments. Normalization by the control runs standard deviation has been used in creating all three indices. As expected, Fig. 9 shows that there are differences in the NAO index between the two simulations and this needs to be considered when assessing the coupled model response. In particular, account must be taken of the fact that the near-surface ocean temperatures will be significantly affected by the atmospheric forcing associated with the state of the NAO in each of the simulations.

The experiments were stopped after 14 years since at this point feedbacks between various effects of the extended period of deep convection in CONV gave rise to drifts in the model ocean state. All results shown are for wintertime means (December–February).

Figure 10 shows the wintertime surface temperature and salinity differences between the sensitivity studies meaned over years 10–14 and the HadCM3 control mean for equivalent years. To some extent these surface fields will be effected by the interannual variability in the NAO forcing of the ocean and for this reason 5-yr means have been used to average out this effect. The sea surface in the subpolar gyre in CONV is generally more saline and warmer than in either INHIB or the control although the peak changes are concentrated in a relatively narrow zonal band. A simple test based on the standard deviation of SST in a 50-yr period of HadCM3 shows that the subpolar gyre warming in CONV is statistically significant, while the cooling in INHIB is within the expected range of the control integration. This is because the two experiments do not act in completely opposite ways, perhaps because the section of HadCM3 used has very little convection in the Labrador Sea and is thus very close to INHIB. This needs to be kept in mind when interpreting these experiments. Results shown later illustrate that the response in CONV is considerably larger than in INHIB. Inclusion of freshwater forcing has led to a slight cooling over the majority of the subpolar gyre and a freshening along surface currents. This could be spurious since the forcing added more freshwater to an already unrealistically stable layer. Although not shown here, sections across the gyre confirm that the fresh layer in subpolar gyre in both INHIB and the control is shallow, rarely reaching deeper than the top 100 m of the water column, and that it allows the presence of a temperature inversion. If the unrealistic stable surface layer is removed the temperature inversion cannot be maintained. From the initial conditions of CONV (and thus INHIB) it is calculated that ∼1°C warming at the surface arises as a direct result of vertically mixing the water column in order to remove the inversion (along 50°N, maximum values of 1.5°C occur in the central Atlantic region). In CONV the freshwater-induced stable surface layer present in the initial conditions (from the control run) is quickly broken down through addition of the Labrador Sea salt forcing. By years 10–14 this breakdown is a major contribution to the gyre warming at the surface, that is, ∼1°C, and is not an effect expected in reality as it arises from breaking down of the overly fresh and unrealistic surface layers in the modeled subpolar gyre. The saline forcing applied throughout CONV prevents sufficient freshwater being available to reestablish this stable surface lid, thus once removed the temperature inversion does not reappear. The ∼1°C warming must be taken into account in all comparisons of the subpolar gyre with the simulation of either INHIB or this period of the HadCM3 control. Note also that in CONV, an increase in surface salinity is seen in the western section of the gyre and along the gyre–gyre boundary or NAC.

In CONV the saline water, arising from the Labrador Sea forcing term, will first appear in the western part of the subpolar gyre. It will then be advected across the basin by the surface NAC on a timescale of a year or so. The SST anomalies associated with the mixing caused by this saline water will also be advected on the same timescale. This process will contribute to the spreading of the warming throughout the subpolar gyre, although the timescale involved is that of the surface current and is therefore relatively short (∼1 yr). In addition there is also considerable variability in SST associated with atmospheric forcing from the NAO. A comparison of the SSTs for INHIB in years 3 and 7 (years of low and high NAOI, respectively, see Fig. 9) and for CONV in years 11 and 13 (also years of low and high NAOI, respectively, see Fig. 9), shows the strong influence of atmospheric forcing on the SST. For example, in year 13 of CONV, when the NAOI is highly positive, SSTs are colder than those in the low NAOI phase, although still considerably warmer than in the control. In fact, over the period of year 5–12, Fig. 9 shows the NAOI in CONV to be consistently lower than in the control or INHIB. This persistent low NAOI state in CONV will also contribute to the surface warming in the subpolar gyre. To further ascertain whether this warming can be attributed to the applied salinity forcing or a combination of surface forcing and wind stress due to NAO forcing, a four-member ensemble of CONV-type runs was completed. Each run begins from a different atmospheric and oceanic state, running for 14 years with forcing identical to that applied in CONV. Although the NAO evolves differently in each experiment the subpolar gyre warming is very similar to that seen in CONV. Further, the temperature and salinity anomalies in each ensemble member are as extensive and of similar magnitudes to the anomalies in CONV. It can thus be asserted that the subpolar gyre warming is not being driven entirely by surface fluxes or by local NAO forcing. In what follows, we are seeking SST signals that stand out above the signal of natural NAO-related variability in the model.

The temperature and salinity differences after 10 years of integration between CONV, INHIB, and the control at 1000 m, approximately the depth of LSW, are shown in Fig. 11, again averaged over years 10–14 (as in Fig. 10). In CONV an east–west temperature dipole structure has formed, the majority of the subpolar gyre cooling while the eastern side of the basin at the same latitude is warming. The same is seen in the salinity field of CONV, a freshening in the western basin while the eastern side has increased in salinity. The north–south dipole, which is also evident in the subpolar gyre, is due to a combination of both changes in the water masses overflowing the Denmark Straits and effects of the Labrador Sea forcing; however, further discussion will not be included here as it is beyond the scope of this paper. There is actually very little change in INHIB at this depth although there is some warming and an increase in salinity in the Labrador Sea. It will be shown later that most differences at this level can be explained in terms of LSW formation changes. One interesting difference between the two sensitivity studies at this depth is what has happened to the north of the Greenland–Scotland ridge system. In CONV the waters in the Norwegian Sea have become warmer and saltier while in INHIB the eastern side of the Greenland–Iceland–Norwegian (GIN) basin has cooled and freshened. This is due to changes in the water masses that are entering this northern basin from the North Atlantic via the Iceland–Scotland ridge. In INHIB the waters are generally fresher and cooler than in the control, and in CONV the opposite is seen. This has an effect on water formation in the GIN Seas and therefore on the properties of the water that overflows the Denmark Straits, as mentioned briefly above.

In order to understand the temperature and salinity difference in the sensitivity studies we need to look at both changes in circulation in each experiment and any changes in water masses. First we look at the current structure. Figures 12a and 12b show the wintertime mean 50-m currents (averaged over years 10–14) for the two experiments. We choose 50 m rather than the surface to ensure that changes are not due to variations in the Ekman drift forced directly by the wind. It can be seen that the Gulf Stream separates from the coast too far north in both cases but, as noted previously, this is a feature of HadCM3. The northward turn seen in the NAC in the central Atlantic in CONV is almost absent in INHIB. Studies by Wright and Gordon (1997) showed that in a previous version of the Hadley Centre model the presence of a northward turn depended on the water mass produced in the Labrador Sea, both in its properties and volume.

The speed of the NAC has increased in CONV, relative to HadCM3, while in INHIB it has weakened (not shown). At 1000 m the enhanced circulation in the subpolar gyre circulation in CONV becomes more obvious, as illustrated in Fig. 13. INHIB shows a fairly weak circulation. Although not included here plots of the currents at 2100 m and deeper reveal that the DWBC has spun up in CONV. This is consistent with the results of Curry et al. (1998) who show that there is decadal variability in the LSW thickness in the western subtropics.

We turn now to look at isopycnal layers to identify whether there are changes to the volume of water masses, as we would expect to see as a result of deep convection, and to identify the extent of any changes. Figure 14 shows annual mean potential density (σ0) at year 10 on a cross section at 50°N from the North American coast to the European continent. We use year 10, rather than a mean over years 10–14, as the position of the “front” is a transient feature. The increased volume of density class 27.9–28.0 in CONV is striking, as is the steepening of the frontal structure at 30°W. There is little change to the structure across the basin in INHIB. This implies that forcing convection has led to a more voluminous water mass (LSW, density 27.9–28.0) relative to both the control and the case with inhibited convection. The frontal structure observed at 30°W is associated with the NAC, the steepening of the front being linked with an intensification of the northward component of the current in this location (see Fig. 12a). The development of the 27.9–28.0 water begins on the western side of the basin as a thickening of the layer. The evolution of this feature can be seen in Fig. 15a, a distance–time section of the thickness of the 27.9–28.0 isopycnal layer across 50°N. For comparison, Fig. 15b shows the standard deviation of the layer thickness variation calculated from 50 years of the control run. The peak changes from year 5 to 10 in CONV (Fig. 15a) are 6 times the control standard deviation and therefore highly significant. A thickening of this layer first appears on the western extreme of the section after approximately four years. The 4-yr delay is consistent with the time taken for the anomaly to advect from the Labrador Sea convecting region. This thickening continues and moves eastward until reaching 30°W after a further four years. Although not shown here the thickening is primarily confined to the southern part of the gyre, changes on the northern side being governed more by changes in overflow water. Figure 15c shows the SST anomalies for the same section. It can be seen that a warming of greater than 2°C appears shortly after the first signs of the isopycnal layer thickening, that is, approximately four years into the sensitivity experiment, and then propagates eastward. Figure 15d shows the latitudinal standard deviation in SST anomalies along this section, calculated using a 50-yr period of the HadCM3 control. It is clear that the initial warming of ∼1°C is within the natural variability of this region. This warming over the first few years may also be a consequence of the removal of the (erroneous) freshwater layer, as described earlier.

Figure 15c shows that SST changes at 50°N behave in a similar way to changes in LSW thickness and that these changes in SST also propagate eastward at this latitude. Figure 16 shows the corresponding plot of the SST anomaly along the NAC for CONV minus the mean from the HadCM3 control. The NAC path is that taken from the control HadCM3 integration as in Fig. 8. After approximately 3–4 yr a large SST anomaly appears to the south of Newfoundland that then propagates northeastward along the NAC. The propagation speed is seen to be similar to that in Fig. 8. For comparison, Fig. 16b shows the control run standard deviation of SST along the path of the NAC. Away from the gyre “meeting point” (marked X), the changes are typically double the control standard deviation. The presence of the warm anomaly upstream of Newfoundland is intriguing since it is hard to see how if could be directly forced by the conditions imposed in these experiments. This anomaly will be discussed in section 6.

b. Steering of the NAC by LSW convection

It has already been demonstrated in Fig. 14 that as the anomalously thickened LSW layer moves eastward across the basin, the east–west density gradient intensifies. By geostrophy this will drive a northward current, effectively an anomalous northward component of the NAC. Analysis of the upper-ocean current field (see Fig. 17) indicates that this is indeed the case. By year 12, the 300-m currents at the subpolar gyre southern boundary are both stronger and broader in CONV as compared with INHIB. The volume transport across a section south from Newfoundland along 55°W shows a 30% increase in the strength of the flow in CONV in comparison with INHIB after 10–15 yr. This increase will, in itself, lead to increased advection of warm water into the Atlantic basin. As expected, the major difference is the strong northward turn exhibited in the CONV currents. This feature is absent in INHIB (see Figs. 17c and 17d). A decomposition of the current into its baroclinic and barotropic parts confirms that this northward component is dominated by the baroclinic component. Figure 17 shows the currents at 300 m for years 7 and 12. The westward propagation of the northward turn in the NAC is evident in CONV. There is a complex dynamical response of the NAC system to the imposed Labrador Sea forcing. Increased Labrador Sea convection leads to a stronger NAC and also to a marked northward turn that itself propagates along the gyre boundary.

In summary, enhanced convection in the Labrador Sea leads to an increased thickness of LSW and greater transport of LSW in the southern part of the modeled subpolar gyre at intermediate depths (around 1000 m). Thickening of the layer moves first southward and then eastward following the modeled pathway of LSW in the gyre. It takes approximately 3–4 yr for the anomalously thickened layer of LSW to reach the pathway of the NAC (where the subtropical and subpolar gyres meet) and then another 4 yr to reach the central Atlantic (30°W). As this water moves across at depth it causes an increase in density gradient both north–south and east–west. Through geostrophy this in turn leads to both a movement north (via an intensification of the northward component of the current) and strengthening of the NAC. The surface manifestations of this effect are the anomalies in SST that propagate along the southern edge of the subpolar gyre, that is, along the mean path of the NAC. Further, the timescale of the eastward propagation of this effect is set by the advective timescale of the anomalous LSW at intermediate depths rather than the faster surface currents.

c. Cyclic forcing of Labrador Sea convection

As demonstrated above, changes in deep convection in the Labrador Sea have many impacts but, most importantly for this study, a sustained period of convection can have an effect on both the position and strength of the NAC. Sutton and Allen (1997) have also shown that when the NAO was at a minimum in the late 1960s–early 1970s, a cold anomaly could be seen propagating across the North Atlantic (see Fig. 4). Conversely a warm anomaly was seen as the NAO moved into a predominantly positive phase in the early 1990s. In the coupled model experiments described in the previous section we made no attempt to control the NAO forcing on the ocean but in forcing deep convection we mimic one aspect of the oceans response. From the observational record of LSW temperatures there emerges a signal with a period of approximately 14 yr, although this is based on a relatively short record that also shows longer timescale variability (Curry et al. 1998). This is indicating an enhancing and then inhibiting of convection in a 14-yr sequence. In a final sensitivity study we aim to recreate this pattern of forcing in an idealized way.

In the experiments described previously the SST response due to deep convection was identified. We seek to identify the same signals in a set of experiments with cyclic forcing of convection. For simplicity we apply sinusoidal freshwater/salt forcing. The range of forcing is such that when it is at a maximum the forcing is the same as in CONV, while at it's minimum the forcing is as in INHIB. In these experiments a cycle time of 14 yr was chosen so that the Labrador Sea experiences convection inhibiting conditions for the first half of the cycle followed by convection enhancing conditions for the later part of the cycle. The forcing cycle is illustrated on Fig. 18. The experiment was run for 42 yr allowing the completion of three cycles.

For the HadCM3 control run we used temperature and salinity between 1000 and 1500 m as an indicator of deep convection in the Labrador Sea. In these experiments, however, we are applying external salt forcing that makes the salinity at this depth harder to interpret in terms of convection. For this reason we concentrate on the mean temperature at this depth (shown in Fig. 18) in order to assess the success of the forcing in controlling convection. If it is doing so we would expect to see temperature at these depths varying in the opposite phase to the forcing. Thus we would hope to see an increase in mean temperature indicating a lack of deep convection, followed by a decrease in temperature as deep convection is enhanced. This cycle should be repeated in each 14-yr period. Figure 18 shows this to be broadly true.

The coupled model shows multiple signals in SST on a range of timescales and so some filtering of the data is required to clearly extract the signal of interest. Figure 19 shows power spectra of SST in the region of high variance, where the NAC turns northward (43°N, 41°–47°W), for the HadCM3 control and the cyclic forcing experiment. There is clearly increased power at the 14-yr timescale in the SST variability of the cyclically forced run in comparison with the control. To project out the signature of this SST variability along the path of the NAC, Fig. 20 shows the SST anomalies from this cyclic forcing simulation along the mean pathway of the control NAC. To extract the decadal signal the data have been filtered, removing signals of less than 7 yr and greater than 20 yr. The higher frequencies are removed to separate out short-timescale variability. On longer timescales there are interactions of the 14-yr cycle with production of deep water in the Greenland Sea. The advection of NAC water into the Nordic Seas is important in determining the watermass properties of the deep water in this region. This interaction manifests in a longer timescale signal in the Greenland–Scotland ridge overflow water and the subpolar gyre. Although these interactions are very interesting, they are beyond the scope of this paper. A pulsating 14-yr SST anomaly signal is very evident at the point where the NAC turns north. The anomaly then propagates to the east with decreasing amplitude. There also appears to be some signal that is coherent upstream of Newfoundland.

A further sensitivity experiment was also performed in which the phase of the cyclic forcing was shifted by 180°, so that in the first half cycle the convection was enhanced. The propagating anomalies along the NAC were not so evident. However, when differences from the control were calculated a clear signal is seen. Indeed, when compared with the first experiment the SST variations along the NAC are 180° out of phase as expected. In both experiments there is an approximately 4-yr lag between the cyclic forcing in the Labrador Sea and the SST response in the region where the NAC turns northward.

These experiments give a clear illustration of how decadal variations of deep convection in the Labrador Sea can lead to corresponding SST variations propagating along the path of the NAC.

6. Discussion

There is a major unresolved issue in the results presented here; namely, that this mechanism can only explain SST anomalies along the northern part of the Gulf Stream/NAC pathway, it does not explain why an anomaly should appear to propagate from Florida to Iceland. The unsmoothed data shown in Fig. 4 indicates that the propagation of an anomaly along the whole path is actually quite a rare event, although in the late 1940s and late 1960s there is evidence of anomalies developing simultaneously along the NAC between Florida and Newfoundland that then continue to propagate along the NAC beyond Newfoundland. The simultaneous SST changes between Florida and Newfoundland in these two time periods can most easily be explained as a local response to the atmospheric NAO forcing. Figure 1a shows that in the late 1940s the NAO was in a relative positive phase whereas in the late 1960s it was at a relative minimum. The SST response associated with the NAO is the tripole SST pattern (Deser and Blackmon 1993). The Gulf Stream/NAC path is such that between Florida and Newfoundland it cuts through one pole of the tripole. In the NAO negative phase this western subtropical part of the SST tripole is cold and in the NAO positive phase it is warm. The direct forcing of the NAO can therefore explain the warm anomalies developing simultaneously along the Gulf Stream path between Florida and Newfoundland in the late 1940s (NAO positive). It can also explain the cold anomalies developing in this region in the late 1960s (NAO negative). Note also that Figs. 6a and 8 show that a similar relationship holds for this region in the model for years 560–565 (cold SST, low NAOI) and 578–583 (warm SST, high NAOI).

An overall plausible mechanism for the propagation of these anomalies is therefore as follows. Because of direct forcing by the NAO, during a period of sustained positive (negative) NAOI the SST anomaly between Florida and Newfoundland will be positive (negative). Sustained positive (negative) NAOI also leads to increased (decreased) Labrador Sea convection. This, in turn, causes an increased (decreased) LSW water mass to circulate along the southern edge of the subpolar gyre at intermediate depths. The movement of this water mass repositions the NAC as it moves eastward and has a signature at the surface of an SST anomaly propagating along the time mean position of the NAC (with the timescale set by the movement at depth of LSW). Increased Labrador Sea convection leads to positive SST anomalies off Newfoundland and so the sign of the anomalies is the same as that caused by direct NAO forcing upstream in the subtropics. This can explain the coherence of the SST signal along the whole path from Florida to Iceland with the whole process being governed by the NAO.

7. Conclusions

The latest version of the Hadley Centre coupled climate model, HadCM3, shows considerable realism in many aspects of decadal variability in the North Atlantic. The relationship between the NAO and Labrador Sea convection (Dickson et al. 1996) is seen in HadCM3 as are the relatively slow moving SST anomalies that appear to propagate along the pathway of the NAC (Sutton and Allen 1997).

The series of sensitivity experiments presented here are based on HadCM3 in which Labrador Sea convection was controlled by addition of a localized salinity forcing. In the first two experiments convection was either continually enhanced or inhibited. The resulting changes in the thickness of the LSW layer could be traced around the subpolar gyre. Repeated winters with deep convection in the Labrador Sea leads to an anomalously thick LSW layer, which in turn gives rise to increases in isopycnal gradients at the anomalously thickened layer. The currents that result from this, by geostrophy, act to both intensify the NAC and to increase the northward turn. Both effects have a warming effect on SSTs along the mean pathway of the NAC. As the anomalously thickened layer moves eastward in the subpolar gyre so the anomalous currents move to the east, an effect that appears at the surface as a propagation of SST anomalies. The key factor then in the relatively slow speed of these anomalies is that they are not related to surface currents but to the advection at depth of the thickened LSW layer.

This mechanism is further illustrated by the results of experiments in which cyclic forcing was applied such that the ocean received convection inhibiting and then enhancing conditions in regular cycles of 14 yr. The result (seen after filtering the data) is a pattern of alternating warm/cold anomalies that propagate along the NAC pathway from Newfoundland toward Iceland.

The results described in this paper go some way in providing an understanding of the origin of SST anomalies propagating along the path of the NAC, at least in the model. The mechanism for the slow propagation speed of surface anomalies along the NAC proposed here is based on the results of a modeling study. To establish whether it is important in reality will require a detailed analysis of historical subsurface data.

Acknowledgments

We would like to acknowledge the useful comments and advice given during the preparation of this paper by our colleagues Howard Cattle, Anne Pardaens, and Peili Wu. We also acknowledge the helpful comments of the reviewers. This work was funded by the Department of the Environment, Transport, and Regions Climate Prediction Programme and the Public Meteorological Service Research Programme.

REFERENCES

  • Appenzeller, C., , T. F. Stocker, , and M. Anklin, 1998: North Atlantic oscillation dynamics recorded in Greenland ice cores. Science, 282 , 446449.

    • Search Google Scholar
    • Export Citation
  • Basnett, T. A., , and D. E. Parker, 1997: Development of the Global Mean Sea Level Pressure Data Set GMSLP2. Hadley Centre Climate Research Tech. Note 79, Met Office, Bracknell, Berkshire, United Kingdom.

    • Search Google Scholar
    • Export Citation
  • Cattle, H., , and J. Crossley, 1995: Modelling Arctic climate change. Philos. Trans. Roy. Soc. London, A352 , 201213.

  • Collins, M., , S. F. B. Tett, , and C. Cooper, 2001: The internal climate variability of HadCM3, a version of the Hadley Centre coupled model without flux adjustments. Climate Dyn, 17 , 6181.

    • Search Google Scholar
    • Export Citation
  • Cook, E. R., , R. D. D'Arrigo, , and K. R. Briffa, 1998: A reconstruction of the North Atlantic oscillation using tree-ring chronologies from North America and Europe. Holocene, 8 , 917.

    • Search Google Scholar
    • Export Citation
  • Cox, M. D., 1984: A primitive equation, 3 dimensional model of the ocean. GFDL Ocean Group Tech. Rep. No. 1, Princeton, NJ, 143 pp.

  • Curry, R., , and M. S. McCartney, 2001: Ocean gyre circulation changes associated with the North Atlantic oscillation. J. Phys. Oceanogr, 31 , 33743400.

    • Search Google Scholar
    • Export Citation
  • Curry, R., , M. S. McCartney, , and T. M. Joyce, 1998: Oceanic transport of subpolar climate signals to mid depth subtropical waters. Nature, 391 , 575577.

    • Search Google Scholar
    • Export Citation
  • Czaja, A., , and C. Frankignoul, 1999: Influence of the North Atlantic SST on the atmospheric circulation. Geophys. Res. Lett, 26 , 29692972.

    • Search Google Scholar
    • Export Citation
  • Deser, C., , and M. L. Blackmon, 1993: Surface climate variations over the North Atlantic Ocean during winter: 1900–1989. J. Climate, 6 , 17431753.

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

    • Search Google Scholar
    • Export Citation
  • Gordon, C., , C. Cooper, , C. A. Senior, , H. T. Banks, , J. M. Gregory, , T. C. Johns, , J. F. B. Mitchell, , and R. A. Wood, 2000: The simulation of SST, sea ice extents and ocean heat transports in a version of the Hadley Centre coupled model without flux adjustments. Climate Dyn, 16 , 147168.

    • Search Google Scholar
    • Export Citation
  • Hall, M. M., , and H. L. Bryden, 1982: Direct estimates and mechanisms of ocean heat transport. Deep-Sea Res, 29 , 339359.

  • Hurrell, J. W., 1995: Decadal trends in the North Atlantic oscillation: Regional temperatures and precipitation. Science, 269 , 676679.

    • Search Google Scholar
    • Export Citation
  • Latif, M., 1998: Dynamics of interdecadal variability in coupled ocean–atmosphere models. J. Climate, 11 , 602624.

  • Levitus, S., , and T. P. Boyer, 1994: Temperature. Vol. 4, World Ocean Atlas 1994, NOAA Atlas NESDIS 4, 117 pp.

  • Levitus, S., , R. Burgett, , and T. P. Boyer, 1994: Salinity. Vol. 3, World Ocean Atlas 1994, NOAA Atlas NESDIS 3, 99 pp.

  • Macdonald, A. M., , and C. Wunsch, 1996: Oceanic estimates of global ocean heat transport. U.S. WOCE Rep. 8, 28–30.

  • Parker, D. E., , C. K. Folland, , and M. Jackson, 1995: Marine surface temperature: observed variations and data requirements. Climatic Change, 31 , 559600.

    • Search Google Scholar
    • Export Citation
  • Pope, V. D., , M. Gallani, , P. R. Rowntree, , and R. A. Stratton, 2000: The impact of new physical parameterisations in the Hadley Centre climate model—HadAM3. Climate Dyn, 16 , 123146.

    • Search Google Scholar
    • Export Citation
  • Rayner, N. A., , E. B. Horton, , D. E. Parker, , C. K. Folland, , and R. B. Hackett, 1996: Global Sea-Ice and Sea Surface Temperature data set, 1903–1994. Version 2.2, Hadley Centre for Climate Prediction and Research Tech. Note 74, Met Office, Bracknell, Berkshire, United Kingdom.

    • Search Google Scholar
    • Export Citation
  • Roberts, M. J., , R. Marsh, , A. L. New, , and R. A. Wood, 1996: An intercomparison of a Bryan–Cox-type ocean model and an isopycnic ocean model. Part I: The subpolar gyre and high-latitude processes. J. Phys. Oceanogr, 26 , 14951527.

    • Search Google Scholar
    • Export Citation
  • Rodwell, M. J., , and C. K. Folland, 2001: Atlantic air–sea interaction and seasonal predictability. Quart. J. Roy. Meteor. Soc., in press.

    • Search Google Scholar
    • Export Citation
  • Rodwell, M. J., , D. P. Rowell, , and C. K. Folland, 1999: Oceanic forcing of the winter North Atlantic oscillation and European climate. Nature, 398 , 320323.

    • Search Google Scholar
    • Export Citation
  • Seager, R., , Y. Kushnir, , M. Visbeck, , N. Naik, , J. Miller, , G. Krahmann, , and H. Cullen, 2000: Causes of Atlantic Ocean climate variability between 1958 and 1998. J. Climate, 13 , 28452862.

    • Search Google Scholar
    • Export Citation
  • Semtner, A. J., 1976: A model for the thermodynamic growth of sea ice in numerical investigations of climate. J. Phys. Oceanogr, 6 , 379389.

    • Search Google Scholar
    • Export Citation
  • Sutton, R. T., , and M. R. Allen, 1997: Decadal predictability of North Atlantic sea surface temperatures and climate. Nature, 388 , 563567.

    • Search Google Scholar
    • Export Citation
  • Wood, R. A., , A. B. Keen, , J. F. B. Mitchell, , and J. M. Gregory, 1999:: Localised collapse of the thermohaline circulation in a climate model with increasing atmospheric CO2. Nature, 399 , 572575.

    • Search Google Scholar
    • Export Citation
  • Wright, D. K., , and C. Gordon, 1997: The positioning of the North Atlantic Current. Ocean Applications Tech. Note 14, Met Office, Bracknell, Berkshire, United Kingdom.

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

(a) Observed NAO index based on normalized wintertime (DJF) sea level pressure differences between Lisbon (Portugal) and Stykkisholmur (Iceland), defined and smoothed as in Hurrell (1995). Heavy line is the smoothed index. (b) Composite observed wintertime mean sea level pressure pattern, calculated as the mean of years NAOI > 1 minus the mean of years when NAOI < 1. Both figures calculated using the GMSLP (Basnett and Parker 1997) dataset

Citation: Journal of Climate 15, 1; 10.1175/1520-0442(2002)015<0045:NAODVI>2.0.CO;2

Fig. 2.
Fig. 2.

North Atlantic sea surface temperature (°C): (a) HadCM3 control and (b) GISST 2.2 observed annual mean climatology. (c) HadCM3 50-m currents (cm s−1) (only currents above 4 cm s−1 are plotted), (d) 670-m currents (only currents above 1 cm s−1 are plotted), and (e) 2100-m currents (only currents above 1 cm s−1 are plotted). Model data based on annual mean of years 540–639

Citation: Journal of Climate 15, 1; 10.1175/1520-0442(2002)015<0045:NAODVI>2.0.CO;2

Fig. 3.
Fig. 3.

Standard deviation of SST (°C) from (a) MOHSST (Parker et al. 1995) observed dataset, 1948–94, and (b) the HadCM3 control, years 540–639. Note the model data have been masked after calculating standard deviations for ease of comparison with the observational data

Citation: Journal of Climate 15, 1; 10.1175/1520-0442(2002)015<0045:NAODVI>2.0.CO;2

Fig. 4.
Fig. 4.

Observed wintertime (DJF) sea surface temperature anomalies (°C) along the climatological path of the NAC, as in Sutton and Allen (1997) Fig. 2a, but calculated using GISST version 3.0 dataset. No smoothing applied. Contour interval is 0.4°C and differences greater than 0.2°C are shaded

Citation: Journal of Climate 15, 1; 10.1175/1520-0442(2002)015<0045:NAODVI>2.0.CO;2

Fig. 5.
Fig. 5.

NAO index for the first 1000 yr of the HadCM3 control experiment, calculated using normalized wintertime (DJF) sea level pressure differences between Lisbon (Portugal) and Stykkisholmur (Iceland), defined and smoothed as in Hurrell (1995). Heavy line is the smoothed index

Citation: Journal of Climate 15, 1; 10.1175/1520-0442(2002)015<0045:NAODVI>2.0.CO;2

Fig. 6.
Fig. 6.

(a) NAO index, as in Fig. 5, for year 540–639 of the HadCM3 control run. (b) Composite wintertime mean sea level pressure for year 540–639 of the HadCM3 control, calculated as the mean of years NAOI > 1 minus the mean of years when NAOI < 1

Citation: Journal of Climate 15, 1; 10.1175/1520-0442(2002)015<0045:NAODVI>2.0.CO;2

Fig. 7.
Fig. 7.

Mean temperature (°C), solid line, and salinity (psu), dotted line, in the Labrador Sea between 1000 and 1500 m. Annual mean data, years 540–639 of the HadCM3 control simulation. The smoothed NAOI from Fig. 6 is also shown for ease of comparisons as the dashed line

Citation: Journal of Climate 15, 1; 10.1175/1520-0442(2002)015<0045:NAODVI>2.0.CO;2

Fig. 8.
Fig. 8.

Model wintertime (DJF) sea surface temperature (°C) anomalies, years 540–593, (relative to 540–639 mean) along the mean path of the model NAC. The “X” marks the region in which the subpolar and subtropical gyres meet (off Newfoundland). No smoothing has been applied. Contour interval is 1.0°C and differences greater than 0.5°C are shaded

Citation: Journal of Climate 15, 1; 10.1175/1520-0442(2002)015<0045:NAODVI>2.0.CO;2

Fig. 9.
Fig. 9.

Unsmoothed NAO index for control (solid line) as in Fig. 5. CONV (dashed line) and INHIB (dotted line)

Citation: Journal of Climate 15, 1; 10.1175/1520-0442(2002)015<0045:NAODVI>2.0.CO;2

Fig. 10.
Fig. 10.

Sea surface temperature difference (°C) for (a) CONV minus HadCM3 and (b) INHIB minus HadCM3. Contour interval is 1.0°C and differences greater than 1.0°C are shaded. Sea surface salinity difference (psu) for (c) CONV minus HadCM3, and (d) INHIB minus HadCM3. Contour interval is 0.5 psu and differences greater than 0.5 psu are shaded. Wintertime mean (DJF) of years 10–14 of sensitivity studies, equivalent years for the HadCM3 control. The region over which salt forcing is applied is shown in (a) as a rectangular box

Citation: Journal of Climate 15, 1; 10.1175/1520-0442(2002)015<0045:NAODVI>2.0.CO;2

Fig. 11.
Fig. 11.

1000-m temperature difference (°C) for (a) CONV minus HadCM3 and (b) INHIB minus HadCM3. Contour interval is 0.4°C and differences greater than 0.2°C are shaded. 1000-m salinity difference (psu) for (c) CONV minus HadCM3 and (d) INHIB minus HadCM3. Contour interval is 0.04 psu and differences greater than 0.02 psu are shaded. Wintertime mean (DJF) of years 10–14 of sensitivity studies, equivalent years for the HadCM3 control

Citation: Journal of Climate 15, 1; 10.1175/1520-0442(2002)015<0045:NAODVI>2.0.CO;2

Fig. 12.
Fig. 12.

50-m currents (cm s−1) for (a) CONV and (b) INHIB. Both are wintertime (DJF) averaged over years 10–14. Only current strengths above 4 cm s−1 are plotted

Citation: Journal of Climate 15, 1; 10.1175/1520-0442(2002)015<0045:NAODVI>2.0.CO;2

Fig. 13.
Fig. 13.

1000-m currents (cm s−1) for (a) CONV and (b) INHIB. Both are wintertime (DJF) averaged over years 10–14. Only current strengths above 1.5 cm s−1 are plotted

Citation: Journal of Climate 15, 1; 10.1175/1520-0442(2002)015<0045:NAODVI>2.0.CO;2

Fig. 14.
Fig. 14.

Potential density (σ0) cross section at 50°N. Wintertime mean, year-10 (a) CONV, (b) INHIB, and year-10 equivalent (c) HadCM3 control

Citation: Journal of Climate 15, 1; 10.1175/1520-0442(2002)015<0045:NAODVI>2.0.CO;2

Fig. 15.
Fig. 15.

(a) Thickness of isopycnal layer (27.9 < σ0 < 28.0) at 50°N from CONV. Wintertime means (DJF), (b) latitudinal standard deviation (m) of this layer thickness at 50°N in HadCM3, (c) SST anomalies (°C) at 50°N (relative to 50-yr wintertime mean of HadCM3, temperature anomalies above 2°C are shaded), and (d) latitudinal standard deviation of SST anomaly at 50°N from HadCM3 (°C). All standard deviations are calculated for wintertime means over the same 50-yr period used in calculating the anomalies

Citation: Journal of Climate 15, 1; 10.1175/1520-0442(2002)015<0045:NAODVI>2.0.CO;2

Fig. 16.
Fig. 16.

(a) Wintertime mean SST difference (°C) between CONV and HadCM3 plotted along the pathway of the mean HadCM3 NAC. The “X” marks the region in which the subpolar and subtropical gyres meet (off Newfoundland). Contour interval is 1.0°C, negative contours are dotted, and differences greater than 2.0°C are shaded. (b) Standard deviation of wintertime SST along the pathway (°C), calculated using the same 50-yr period using to derive anomalies

Citation: Journal of Climate 15, 1; 10.1175/1520-0442(2002)015<0045:NAODVI>2.0.CO;2

Fig. 17.
Fig. 17.

Wintertime (DJF) mean currents at 300 m (cm s−1) for (a) CONV and (b) INHIB, year 7, and for (c) CONV and (d) INHIB, year 12. Only current strengths above 4.0 cm s−1 are plotted

Citation: Journal of Climate 15, 1; 10.1175/1520-0442(2002)015<0045:NAODVI>2.0.CO;2

Fig. 18.
Fig. 18.

Modeled potential temperature (°C) of LSW, averaged between 1000 and 1500 m (solid line), in the cyclic forcing experiment. The salinity that was relaxed toward throughout the experiment, in the forcing region, is indicated by the dotted line. Vertical dashed lines mark the start of each new cycle of forcing

Citation: Journal of Climate 15, 1; 10.1175/1520-0442(2002)015<0045:NAODVI>2.0.CO;2

Fig. 19.
Fig. 19.

Power spectra of SST anomaly on a section of the NAC (43°N, 41°–47°W). The solid line is the control experiment, dashed line the cyclic forcing experiment

Citation: Journal of Climate 15, 1; 10.1175/1520-0442(2002)015<0045:NAODVI>2.0.CO;2

Fig. 20.
Fig. 20.

Cyclic forcing experiment, detrended and filtered (retaining 7–20 yr) wintertime SST anomalies (°C) along the mean path of the modeled NAC (taken from HadCM3 control). The “X” marks the region in which the subpolar and subtropical gyres meet (off Newfoundland). Contour interval is 0.5°C, with 0.25°C contour added, and differences greater than 0.25°C are shaded

Citation: Journal of Climate 15, 1; 10.1175/1520-0442(2002)015<0045:NAODVI>2.0.CO;2

Save