The sensitivity of sea ice to the temperature of inflowing Atlantic water across the Greenland–Scotland Ridge is investigated using an eddy-resolving configuration of the Massachusetts Institute of Technology General Circulation Model with idealized topography. During the last glacial period, when climate on Greenland is known to have been extremely unstable, sea ice is thought to have covered the Nordic seas. The dramatic excursions in climate during this period, seen as large abrupt warming events on Greenland and known as Dansgaard–Oeschger (DO) events, are proposed to have been caused by a rapid retreat of Nordic seas sea ice. Here, we show that a full sea ice cover and Arctic-like stratification can exist in the Nordic seas given a sufficiently cold Atlantic inflow and corresponding low transport of heat across the Greenland–Scotland Ridge. Once sea ice is established, continued sea ice formation and melt efficiently freshens the surface ocean and makes the deeper layers more saline. This creates a strong salinity stratification in the Nordic seas, similar to today’s Arctic Ocean, with a cold fresh surface layer protecting the overlying sea ice from the warm Atlantic water below. There is a nonlinear response in Nordic seas sea ice to Atlantic water temperature with simulated large abrupt changes in sea ice given small changes in inflowing temperature. This suggests that the DO events were more likely to have occurred during periods of reduced warm Atlantic water inflow to the Nordic seas.
The forcing mechanisms behind the abrupt climate changes occurring repeatedly during the last ice age, the Dansgaard–Oeschger (DO) events, are still debated. The DO events were first identified in ice cores on Greenland, where the events are characterized by an abrupt warming of 10° ± 5°C in a few decades (Johnsen et al. 1992; Dansgaard et al. 1993; North Greenland Ice Core Project members 2004). During the glacial period, the climate of Greenland alternated between cold stadial and warm interstadial conditions with a period of roughly 1500 years (Grootes and Stuiver 1997). Further climate reconstructions suggested a global extent [see Voelker (2002) and Rahmstorf (2002) for an overview] with, for example, warmer and wetter climate in Europe coinciding with interstadial conditions on Greenland.
Climate reconstructions show large variations in the North Atlantic during the same time period with, for example, sea surface temperature changes of 10°C (Sánchez Goñi et al. 2008) and movements of ocean fronts (Eynaud et al. 2009; Voelker and de Abreu 2013; Rasmussen et al. 2016) occurring with the Greenland stadial–interstadial cycles. Originally, a change in the Atlantic meridional overturning circulation (AMOC) was proposed as a mechanism explaining the millennial-scale climate variability (Broecker et al. 1990; Tziperman 1997; Marotzke 2000; Ganopolski and Rahmstorf 2001). However, it is not clear how a change in AMOC alone can initiate a 10°C warming on Greenland. In recent years, the agent for the warming on Greenland is believed to be abrupt reductions in sea ice cover (Broecker 2000; Gildor and Tziperman 2003; Masson-Delmotte et al. 2005), in particular over the Nordic seas (Li et al. 2005; Dokken et al. 2013). During stadial conditions on Greenland, the Nordic seas are hypothesized to be fully sea ice covered, thereby insulating the relatively warm ocean from the cold atmosphere above. An abrupt reduction in sea ice cover leads to a rapid release of oceanic heat to the atmosphere. In particular, model studies show a 7°C warming on Greenland resulting from sea ice loss in the Nordic seas (Li et al. 2005, 2010). Li et al. (2005) also show that the reconstructed changes in snow accumulation on Greenland during DO events can be explained by abrupt reductions in sea ice cover.
An increasing number of climate reconstructions from marine sediment cores support the existence of a temporal sea ice cover in the Nordic seas during the last glacial period (Dokken et al. 2013; Hoff et al. 2016). In particular, Dokken and Jansen (1999), Rasmussen and Thomsen (2004), Dokken et al. (2013), and Ezat et al. (2014) show that the hydrography of the eastern Nordic seas during stadial conditions resembles the Arctic Ocean today, with a warm subsurface below a cold and fresh surface layer, which is a stratification that may allow for the presence of a sea ice cover (Jensen et al. 2016). During warm interstadials, the Nordic seas are shown to be less stratified, with warm Atlantic water close to the surface. Recently, direct evidence for the presence of sea ice at the same sites is emerging; for example, sea ice biomarkers have been found in oceanic sediments (Hoff et al. 2016).
Today, the eastern Nordic seas are free of sea ice, and the surface ocean is significantly warmer than other oceans at similar latitudes (Hansen and Østerhus 2000). Warm Atlantic water enters the Nordic seas over the Greenland–Scotland Ridge (GSR) and the Norwegian Atlantic Current carries heat and salt toward the Arctic Ocean along the coast of Norway (Orvik and Niiler 2002; see Fig. 1 herein). The result is an eastern Nordic seas with a temperature of approximately 8°C, even in winter (Hansen and Østerhus 2000; Seidov et al. 2013, 2015). Heat is continuously lost to the atmosphere as the current travels northward. However, climate reconstructions of the last glacial period show a temporal, seasonal sea ice cover off the coast of Norway (Hoff et al. 2016). There must have been changes to the hydrography and vertical stratification of the Nordic seas to allow for the existence of sea ice here. Either the incoming Atlantic water cooled substantially, the volume of warm Atlantic inflow was reduced, or the Atlantic water entered the Nordic seas at greater depths.
Here we study the existence and stability of sea ice in the Nordic seas by applying an idealized eddy-resolving configuration of the Massachusetts Institute of Technology General Circulation Model (MITgcm). A similar setup without sea ice has been shown to reproduce most of the characteristic of the modern circulation in the Nordic seas (Walin et al. 2004; Spall 2004, 2011, 2012); in particular, both the observed barotropic and baroclinic part of the boundary current (Orvik et al. 2001) were reproduced by Walin et al. (2004). The warm cyclonic boundary current in the Nordic seas exists because of topography and surface buoyancy forcing. Wind is shown to have little impact on the exchange between the subpolar Atlantic and the Nordic seas (Spall 2011). Eddies, on the other hand, carry heat both across the GSR and into the interior of the Nordic seas and are critical for a realistic representation of the heat exchanges in the area. A coupled ocean–sea ice setup of this model has been applied to study an Arctic Ocean–like domain (Spall 2013). This version is used here to study the Nordic seas and the potential for abrupt changes in sea ice during cold glacial climates.
Our study suggests that a full sea ice cover can exist in the Nordic seas given a reduction in the poleward heat transport across the Greenland–Scotland Ridge. Details of the model are given in section 2, while the results of the experiments and the stability of a Nordic seas sea ice cover are assessed in section 3. The implications of the results for our understanding of the DO events are discussed in section 4.
2. An idealized numerical model of the Nordic seas
We apply an idealized eddy-resolving version of the MITgcm to study the existence of sea ice in the Nordic seas. The model configuration with an idealized Nordic seas basin is shown in Fig. 2. There is no interactive atmosphere; instead, sea surface temperatures are restored toward constant atmospheric temperatures, with a maximum of 10°C in the south and a minimum of −20°C in the north (Fig. 2). Note that there are no time variations in atmospheric forcing; we are in essence studying a constant winter. To keep it simple, precipitation and wind are not included. The reader is referred to Spall (2011, 2012), who found that wind forcing has a minor impact on the heat exchange between the basins compared to buoyancy forcing. The bathymetry consists of sloping sides, with a maximum depth of 2000 m and a 1000-m-deep ridge. The domain is divided into two subdomains, one south and one north of the ridge. The basin north of the ridge is hereafter referred to as the Nordic seas, the focus area of this study.
The horizontal resolution is 5 km, and there are 30 vertical layers; the upper 20 layers are 50 m, and the deepest 10 layers are 100 m thick. We apply a realistic equation of state; the density is calculated with the Jackett and McDougall (1995) formula, a constant Coriolis parameter, f = 1.2 × 10−4 s−1, and vertical diffusivity and viscosity of 1 × 10−5 m2 s−1. Convection is parameterized with implicit vertical diffusion; the diffusivity increases to 1000 m2 s−1 for statically unstable conditions. Horizontal viscosity is parameterized using the Smagorinsky closure (Smagorinsky 1963):
where υs = 2.5 is a nondimensional coefficient, Δ is the horizontal grid spacing, and u and υ are the zonal and meridional velocities, respectively, with subscripts x and y indicating partial differentiation. Typical values are 30 m2 s−1 for the boundary current region. Temperature and salinity are advected using a third-order flux-limiting scheme.
A new feature of this study is that the domain is coupled to a zero-layer thermodynamic sea ice model with viscous-plastic dynamics (Hibler 1980; Zhang and Hibler 1997; Losch et al. 2010). The sea ice model has been included in an Arctic Ocean–like setup where many physical aspects of the Arctic circulation are represented (Spall 2013) but has not been applied to the Nordic seas domain. To mimic river runoff, Spall (2013) included a virtual freshwater forcing that is not included in this study. We are thus studying a sea ice cover in the absence of external freshwater supplies. The sea ice is assumed not to store any heat, there is no snow, and the salinity of the ice is set to 0 psu. Atmospheric restoring is deactivated over sea ice; instead a constant downward longwave radiation FLW↓ of 60 W m−2 is specified where there is sea ice present. The temperature of the sea ice surface Ts is solved to balance FLW↓ + Fc(Ts) − FLW↑(Ts) = 0, where Fc is the heat conduction through the sea ice, and FLW↑ is the upward longwave radiation.
We assume an infinite amount of warm Atlantic water present in the south by restoring the southern boundary to constant temperatures (between 1° and 8°C) and salinities (35 psu) with a restoring strength of 40 W m−2 C−1 and a time scale of 1 month. The model is initialized from rest with a constant salinity of 35 psu, and a temperature corresponding to the restoring temperature at all depths.
We perform eight main simulations where the restoring temperature in the south (hereafter referred to as the Atlantic water temperature or TA) varies between 1° and 8°C. The runs are named XXDEG, where XX is the restoring temperature in degrees Celsius (Table 1). The 4.5DEG–8DEG simulations are run for 60 years, the 1DEG–3DEG simulations are run for 120 years, and the 4DEG experiment that exhibits transient changes in sea ice volume is run for 200 years. All figures (if not specified otherwise) are time averages over 10 years.
First, we investigate the circulation of the Nordic seas domain and the features common to all experiments by using the 6DEG experiment (section 3a) as a reference. Thereafter, we describe the formation of an extensive sea ice cover (section 3b) and its implications for the ocean circulation (section 3c). Theoretical considerations are presented in section 3d.
a. Nordic seas circulation
The buoyancy forcing from atmospheric cooling in the north and warming in the south drives a warm baroclinic inflow to the Nordic seas from the warmer southern domain. The warm water crosses the sill on the western side over the open geostrophic contours, before it travels eastward on the northern sill flank (Fig. 3). Thereafter, the warm water flows cyclonically along the topographic contours around the Nordic seas, similar to the circulation today (Figs. 3 and 4). South of the ridge, the circulation consists of both a cyclonic and an anticyclonic gyre (Fig. 3).
We note that in today’s Nordic seas, the northward flow of warm water across the GSR mainly takes place on the eastern side of the ridge. However, we are interested in the total heat transport into the Nordic seas and the circulation of the Nordic seas itself. As there are no zonal variations in atmospheric temperature, the total heat transport across the GSR is not affected by the location of the inflow. In addition, as the circulation within the Nordic seas–like domain resembles observations, we argue that the inflow on the western side is not critical for our main results. Walin et al. (2004) show a similar circulation in an idealized two-basin setup without wind, while still being able to model the main characteristics of the Nordic seas. We therefore focus on the circulation within the Nordic seas itself and the amount of heat entering the basin.
Exchange of heat across the GSR is secured by the mean flow and eddy heat fluxes, with the latter dominating the total heat transport into the Nordic seas (Fig. 5). The total heat transport is calculated as
where ρ is the density, cp is the heat capacity of seawater, and T is the temperature (referenced to 0°C). The integration is carried out over the full depth H and across the entire sill crest (X = 1000 km). As υ and T can be divided into their mean and residual components, the total heat transport can be divided into that caused by the mean flow and that from eddy heat fluxes (Fig. 5). Eddy heat fluxes contribute two-thirds of the total heat transport across the ridge, and originate from the unstable temperature front across the sill crest. However, in the inflow region (the westernmost 100 km) the mean flow dominates the heat transport [Fig. 5; Eq. (1) with X = 100 km]. Here, warm water is advected northward with the mean flow, except for a cold outflow at depth. Along the rest of the sill crest, eddy heat fluxes carry heat northward and are thus crucial for a correct representation of the net heat exchange.
North of the sill, the main northward flow is confined to a baroclinic eastern boundary current (here defined as the easternmost 100 km and 250–1000 km north of the sill), while the southward flow is both in the upper part of the western boundary current (defined as the westernmost 100 km and 250–1000 km north of the sill), and at depths in the eastern boundary current (Fig. 4). The horizontal extent is limited by the sloping bathymetry. Below sill depth (1000 m) there is a weak anticyclonic boundary current. Walin et al. (2004) also find an anticyclonic vortex in the deep with very little relevance for the surface circulation. There is also a weak barotropic cyclonic gyre off the sloping bathymetry (as seen in oceanographic observations; Orvik et al. 2001).
The interior of the Nordic seas domain is not connected to the southern domain via open geostrophic contours. As a consequence, the net heat transport in the interior is mainly driven by eddy heat fluxes. The unstable baroclinic boundary current sheds off eddies to the interior. However, the cold atmosphere manages to cool down the interior to the freezing point, making the interior the coldest part of the Nordic seas. For all experiments, sea ice is present in the interior of the Nordic seas (Fig. 6) because of the cold atmosphere and limited heat transport into the interior.
b. Sea ice–induced halocline formation
Although the main circulation remains the same for all experiments, the different Atlantic water temperatures lead to two distinctly different states for the Nordic seas: one where the entire domain is covered in sea ice (1DEG–4DEG; Fig. 6) and one where only the interior is sea ice covered while the margins are ice free (4.5DEG–8DEG; Fig. 6). The states are hereafter referred to as COLD and WARM, and represented by the 3DEG and 6DEG experiments, respectively. Within each subset, the differences between experiments are much smaller and more gradual than differences between the two different states.
For the WARM experiments, both the net heat transport and the total sea ice volume change close to linearly with Atlantic water temperatures (Fig. 5). The change in sea ice volume is mainly a result of sea ice thickness changes, but also a result of greater sea ice area: As the warm cyclonic boundary current cools by heat loss to the atmosphere and by eddy fluxes to the interior, the sea ice extends farther out into the warm boundary current. When moving to lower Atlantic water temperatures, the linear decrease in heat transport is mainly the result of a decrease in eddy heat fluxes as the north–south temperature gradient decreases, and a linear decrease in the velocity of the boundary current.
In contrast, changes in the heat transport and sea ice volume are highly nonlinear when transitioning from the WARM to the COLD experiments (Fig. 5), suggesting that feedbacks in response to the decrease in TA are causing the rapid expansion of sea ice. The COLD experiments all have sea ice covering the entire Nordic seas, including the warm boundary current; the volume differences within the COLD experiments are due to sea ice thickness. We use the COLD experiment (3DEG), which has an extensive sea ice cover in the Nordic seas, to investigate how the sea ice is formed. Ice volume reaches a steady state after about 30 years. However, as sea ice area reaches its maximum after only 10 years, we will study the underlying dynamical processes during the first decade.
As the atmosphere cools the Nordic seas domain, the surface ocean becomes denser (Figs. 7a,c). Seeing that there are no freshwater sources and no precipitation added to the domain, the initial salinity is 35 psu everywhere with a freezing point of −1.9°C. This temperature is reached in the north of the domain after a few months. Sea ice formation starts, and brine rejection increases the salinity of the ocean below, leading to vigorous deep convection. As the sea ice thickens, salinity at depth increases. Given that a restoring to 35 psu is applied in the south, part of the excess salt is removed, and the restoring acts as a virtual freshwater supply.
Freshwater is stored in the newly created sea ice. When sea ice meets warm Atlantic water, it melts and releases freshwater back to the surface. As a consequence, enhanced sea ice production acts to stratify the Nordic seas by efficiently freshening the surface, and making the deeper ocean more salty and dense (Fig. 7b). The melting of sea ice lags the freezing in the beginning, but eventually the two balance each other. Sea ice is transported from the regions with active sea ice formation to the regions with sea ice melt. Thus, advection of sea ice by the ocean plays an important role for the stratification, as sea ice is advected into the warm boundary current where it melts and releases freshwater. An additional sensitivity experiment with a purely thermodynamic sea ice model (replacing the dynamic sea ice model including advection by ocean currents) resulted in sea ice existing only in the interior of the Nordic seas for all experiments (also for the COLD experiments). For the dynamic sea ice model, the mean flow dominates the total advection of sea ice, except for periods with little sea ice when eddy fluxes dominate (not shown).
As a consequence of the stratifying effect of sea ice, a halocline emerges for the COLD experiments (Figs. 7 and 8). Sufficient ice is melted to create a protective fresh surface layer, forcing the relatively warm and salty Atlantic water to subduct as it flows north into the basin. The relatively salty and warm Atlantic waters (>0°C), which dominated the surface circulation of the Nordic seas, are now more dense than the fresh surface layer and subduct to approximately 200-m depths (Figs. 7c and 8). Thus, the warm boundary current is still present in the COLD experiments, but it is located below the fresh surface layer (Fig. 4b). The ensuing stratification allows for an expansion of the sea ice cover as the fresh cold layer efficiently separates the sea ice from the warmer Atlantic water below.
The halocline is absent in the boundary current region for the WARM experiments (Fig. 8). The stratification by production and melting of sea ice also occurs in the WARM experiments where the atmosphere cools the surface ocean in the interior of the Nordic seas basin to freezing temperatures. However, the buoyancy flux from freshwater input from melting sea ice is smaller than in the COLD experiments (Fig. 9a), especially in the beginning of the experiment (not shown). In the WARM experiments, there is not enough meltwater present to create a fresh surface layer, and the warm Atlantic water is still at the surface in the boundary current region, preventing sea ice formation (Fig. 4). In addition, as the Atlantic water is colder for the COLD experiments, less freshwater is needed to produce a surface layer lighter than the Atlantic water (section 3d).
In summary, sea ice in the Nordic seas basin forms because of cooling by the atmosphere, and the extent is aided by advection and melting of sea ice, which creates a fresh surface layer. A halocline emerges spontaneously once the temperature of the inflowing Atlantic water is sufficiently reduced.
Interestingly, the 4DEG experiment, which is the warmest of the COLD experiments, exhibits unforced transient changes in sea ice cover (Fig. 10). The experiment shifts repeatedly between the COLD and the WARM state. For this experiment, there is just enough meltwater present to create a fresh surface layer. However, with time the subsurface becomes warmer and eventually lighter than the surface layer. As a result, warm water is mixed up and sea ice melts. The eastern and northern boundary current becomes temporarily ice free.
c. Changes in heat transport and meridional overturning circulation
The presence of an extensive sea ice cover has implications for the ocean circulation. When the boundary current area is covered by sea ice, the release of heat to the atmosphere is reduced (Fig. 9b). Hence, the western boundary current is much warmer when sea ice is present, and the warm boundary current is isolated from the atmosphere by a strong halocline (Figs. 4b and 8a). In 6DEG, the warm water has cooled by approximately 3°C when reaching the western boundary current. In 3DEG, the cooling of the warm boundary current from the eastern margin to the western margin is reduced to approximately 0.3°C. (Fig. 8a). The cooling in 6DEG is mainly attributed to eddy heat fluxes and atmospheric cooling (Fig. 8c). As parts of the temperature–salinity (TS) diagram for 3DEG follows the Gade line (density change resulting from the salinity and temperature decrease associated with sea ice melt; Gade 1979), energy lost to sea ice melt contributes to the cooling in 3DEG (Fig. 8d). Note that below 800-m depth the western boundary is colder in 3DEG than in 6DEG.
Total heat transport into the Nordic seas is low when a full sea ice cover is present (Fig. 5). This is a result of low Atlantic water temperatures but also a consequence of the insulating sea ice cover, which reduces the temperature difference between the inflowing and outflowing water across the sill. As the release of heat to the atmosphere is reduced with a full sea ice cover, heat is recirculated within and out of the Nordic seas. In addition, the reduction in the velocities across the sill crest and in the boundary current (Fig. 4) impacts the total heat transport.
The strength of the Nordic seas meridional overturning circulation (NMOC) decreases when a full sea ice cover is present (Fig. 11). As with the net heat transport, the decrease could either be due to smaller buoyancy forcing caused by a decrease in TA and a consequent smaller north–south temperature gradient, or as a result of a full sea ice cover. To investigate this further, we study the 4DEG experiment with transient changes in sea ice volume. In the 4DEG experiment the NMOC varies between 2 Sv (1 Sv = 106 m3 s−1) in the state with maximum sea ice and 6 Sv in the state with minimum sea ice. As TA is kept constant throughout the experiment, the shift in the strength of the NMOC is due to sea ice changes and not to changes in the inflowing Atlantic water. In contrast, the 5DEG experiment has a maximum NMOC of 6 Sv (Table 1), suggesting that the largest part of the decrease in NMOC from the WARM to the COLD state is due to the presence of a sea ice cover and not to changes in TA (i.e., buoyancy forcing). However, as with the total heat transport, there is a close to linear increase in the overturning with TA for the WARM experiments, showing the importance of buoyancy forcing when sea ice is at a minimum. The linear relationship agrees with the theory developed by Spall (2012), where the circulation scales with the difference between inflowing temperature and the atmospheric temperature.
The effect of a full sea ice cover and reduced heat loss to the atmosphere is a decrease in downwelling and mixed layer depths (Figs. 9 and 12), contributing to the decreasing NMOC. We note that a deep negative overturning cell is present for the COLD experiments, likely a result of increased brine rejection from enhanced sea ice formation. In addition, the temperature of the deep water that exits the Nordic seas across the sill decreases with a full sea ice cover.
d. Theoretical solutions
There is a nonlinear relationship between the Atlantic water temperature and the sea ice volume (Fig. 5). All experiments with TA < 4°C resemble the COLD state, while the experiments with TA > 4°C resemble the WARM state. At 4°C, the experiment shifts between the WARM and the COLD state. The limit between the two states is set by whether sea ice can expand over the boundary current or not, which again depends on the parameters chosen. For example, if the atmospheric temperature forcing increases, the limit between WARM and COLD experiments shifts to lower TA.
1) Atmospheric cooling
The sea ice melts if the ocean heat flux Fw into the ice is larger than the conductive heat flux Fc through the ice; Fc is balanced by the atmospheric fluxes Fatm. In steady state,
where T0 is the temperature of the surface layer, Tf is the freezing temperature of seawater, is a mixing efficiency constant, hi is the sea ice thickness, λ is the thermal conductivity of the sea ice, Ts and Tsk are the surface temperature of the sea ice in degrees Celsius and kelvin, respectively, ε is the emissivity of sea ice, and σ is the Stefan–Boltzmann constant. Equation (3) allows for very small differences between T0 and Tf [O(0.1)°C].
2) Sea surface temperature
Equation (3) can be solved for Ts and hi given T0. In the beginning of the simulations and in the absence of sea ice, the surface temperature of the interior is determined by eddy heat fluxes from the south and the boundary current and by atmospheric cooling. Therefore, T0 can be solved from
where P is the perimeter of the boundary current or the sill crest, A is the area of the region, Tatm is the atmospheric temperature, and Γ is a relaxation constant. Following Visbeck et al. (1996) and Spall (2004, 2012, 2013), we parameterize the eddy heat fluxes as being proportional to the baroclinic velocity in the boundary current V, and the temperature–salinity difference between the interior and boundary current/southern region. Assuming small salinity differences, the eddy heat fluxes can be parameterized as
and T0 solved from
where c is a nondimensional coefficient, g is the gravitational acceleration, L is the width over the sloping topography, and α is the thermal expansion coefficient. Solving Eq. (6) for the area just north of the sill, the balance predicts the surface temperatures for the WARM experiments (Fig. 13). However, the balance does not apply to the cold sea ice–covered state where surface temperatures are close to freezing.
3) Sea ice melt
We also calculate how much sea ice must melt to produce a fresh surface layer with freezing temperatures that is lighter than the Atlantic water for each experiment. First we calculate the highest possible sea surface salinity S0, which gives ρ0 < ρA, where ρA = ρ(SA = 35, TA = TA) and T0 = −1.9°C. Then, given a sea ice salinity of Si, we calculate the fraction of this water, f0, needed to mix with Atlantic water SA to produce water with S = S0:
where ρice is the density of sea ice. As the horizontal areas of the grid cells are the same, the sea ice thickness needed to melt to make the surface layer with a constant depth h1 sufficiently light is
More ice needs to melt to create a light enough surface layer for the WARM experiments compared to the COLD experiments (Fig. 13b). Noticeably, the sea ice thickness that needs to melt for a stable stratification is larger than the mean sea ice thickness of the WARM experiments and so no halocline can be supported in this parameter range.
By applying an idealized setup of the MITgcm we show that a full sea ice cover can exist in a Nordic seas–like domain together with an active Atlantic inflow. However, for a full sea ice cover to be present, the temperature of the Atlantic water needs to be reduced, as well as the heat transport to the Nordic seas. As a consequence, there is a subtle balance between the warm north-flowing Atlantic water, a fresh surface layer, and extensive sea ice formation in the Nordic seas.
The existence of a self-sustained sea ice cover in the Nordic seas is relevant for the large-scale climate variability of the last glacial period where a temporal Nordic seas sea ice cover is hypothesized to govern the abrupt temperature fluctuations recorded in Greenland ice cores. An external freshwater supply is often assumed to be present for a sea ice cover to exist in the Nordic seas. This has typically been attributed to freshwater input from the Laurentide Ice Sheet during stadials (e.g., Broecker et al. 1990; Manabe and Stouffer 1995), enhanced runoff from the Fennoscandian Ice Sheet during interstadials (Dokken et al. 2013), or enhanced ice export from the Arctic Ocean (Tarasov and Peltier 2005). For the first case, runoff is assumed to be largest during cold stadial periods. This is problematic as freshwater is needed the most when the climate state is the least supportive of melting. In addition, episodic freshwater input from the Laurentide Ice Sheet is mainly associated with Heinrich stadials, and not every stadial period.
Here we show that a sea ice cover and Arctic-like stratification can exist without an external freshwater supply. Under sufficiently cold conditions, a halocline, capped by sea ice, emerges spontaneously. The presence of an intermittent freshwater source during the last glacial period is therefore not necessary to explain the cold fresh surface layer that proxy data suggest was present during Greenland stadials. We do note that the Arctic-like stratification also emerges in the presence of external freshwater inputs. Sensitivity studies with a virtual freshwater flux along the eastern margin of the model domain demonstrate that additional freshwater aids in the formation of a halocline. However, the nonlinear transition between the WARM and COLD states still exists; it just moves to slightly higher Atlantic water temperatures.
The unforced halocline formation also has implications for the modern Arctic Ocean. It is debated whether the cold halocline is formed by the addition and advection of cold and salty water produced by sea ice formation on shallow shelves (Aagaard et al. 1981), or by convection in the deep ocean during winter and the addition of freshwater from river runoff and sea ice melt to the surface (Rudels et al. 1996). A halocline is assumed to be a prerequisite for a sea ice cover to exist. Here we show that processes similar to those proposed by Rudels et al. (1996) can, under the present idealized configuration, create a halocline. However, freshwater from river runoff is not needed. Similar to modern oceanographic observations north of Svalbard (Rudels et al. 2005), the Atlantic water cools by heat loss and sea ice melt. The sea ice melt is mainly a seasonal signal (Rudels et al. 2005; Pemberton et al. 2015), but we show that sea ice melt also occurs under cold conditions associated with winter. As cold conditions are crucial for the stability of the halocline, the observed warming of the Atlantic water in recent decades (Polyakov et al. 2005, 2012) might have profound implications for the stability of Arctic sea ice cover.
The highly nonlinear response in sea ice volume to Atlantic water temperatures suggests that large changes in sea ice could occur with small changes within a given range of Atlantic water temperatures. Several studies have shown that the variability of the climate system depends on the background climate state (e.g., Ganopolski and Rahmstorf 2001; Bitz et al. 2007), with less variability in a cold climate than in an intermediate climate. This has typically been related to the recovery rate of the thermohaline circulation after a shutdown resulting from freshwater input (Rahmstorf 1996; Stouffer et al. 2006; Bitz et al. 2007). However, sea ice expansion, a stronger stratification, and additional surface freshwater input through seasonal melt have been shown to affect the recovery rate and hence contribute to a more stable climate in a cold state (Bitz et al. 2007). Also, ice-core records (Dansgaard et al. 1982, 1993) show more variability during intermediate climates; the DO events are present in Marine Isotope Stage 3 (MIS3), which is warmer than the more stable Last Glacial Maximum and colder than the much more stable Holocene. In addition, cold stadials are typically more stable than interstadial periods. In our coldest experiments (1DEG–3DEG), the sea ice cover in the Nordic seas is very stable and the temperature of the Atlantic water must increase by a couple of degrees to trigger a transition into a warm state with significant changes in ice cover. However, this is not the case for the intermediate experiments (e.g., 4DEG), where a small warming of the incoming Atlantic water provides a rapid transition into the warm state with dramatic changes in sea ice cover and heat transport.
This suggests that given the appropriate background climate, such as MIS3 during the last glacial period, a small change in ocean circulation and heat transport into the Nordic seas could have large consequences for the Nordic seas and surrounding climate. This oceanic change could involve larger-scale changes to, for example, the AMOC or more local changes to, for example, the horizontal circulation and the subpolar gyre, which are seen to have large effects on the heat transport into the Nordic seas (Boccaletti et al. 2005; Hátún et al. 2005). For the 4DEG experiment, transient changes in sea ice are even present without a change in the incoming Atlantic water, suggesting that large sea ice variability could also be intrinsic to an intermediate climate state. The observed link between sea ice variability and the background climate state in both models and observations could explain why larger variability is found in the North Atlantic during the intermediate climate state of MIS3 than during colder and warmer periods of the past.
a. Limitations of the model
The seasonal cycle is a key mechanism missing in this study. The seasonal variability of sea ice could prevent abrupt reductions in sea ice cover (Tietsche et al. 2011). In particular, seasonal sea ice would enhance the stratification because of melting and freshwater release during summer, and sea ice production and brine rejection during winter. Therefore, we expect the transition between the COLD and the WARM state of the model to occur at higher Atlantic water temperatures if seasonality were to be included as more sea ice melt would occur. On the other hand, the consequent change in heat content of the ocean could prevent sea ice formation and shift the limit to lower temperatures. A follow up to this study will therefore include a representation of the seasonal cycle in the climate forcing.
In our experiments, the meridional overturning circulation decreases as a consequence of increasing sea ice cover in the Nordic seas. However, we note that this is the local overturning in the Nordic seas (i.e., NMOC) only, and not the full AMOC. There is a constant supply of warm salty water in the south of the domain, which is not sensitive to changes in the NMOC. We are thus not representing the NMOC’s potential impact on the circulation farther south. In reality, the circulation changes in the Nordic seas have the potential to impact the global circulation. However, it is not clear whether changes in the NMOC would affect the AMOC, and this relationship needs to be further investigated. If AMOC weakened, it could lead to a reinforcement of the sea ice cover as a reduced AMOC is associated with a smaller heat transport into the Nordic seas. However, the heat transport by the subpolar gyre could compensate, or also be affected by the sea ice cover resulting from, for example, wind or density changes (Born and Levermann 2010; Born et al. 2015).
For simplicity we have ignored the effects of wind in this study and, as shown by Spall (2011), adding wind forcing representative of the present climate has little effect on the net heat transport to the Nordic seas. However, significant changes in the wind fields are thought to have occurred during parts of the last glacial period when the ice sheets and sea ice were different from today (Löfverström et al. 2014). Evidence showing large changes in the dust content in the ice cores on Greenland between stadial and interstadial transitions suggests significant changes to the atmospheric circulation during MIS3 (Mayewski et al. 1994). In addition to the potential for changes in the wind forcing through time, the impact of wind in general on the sea ice cover could be important and should be considered in further studies.
As discussed, the dynamic component of the sea ice model is required in order to advect sea ice out of the interior of the basin, and produce a full sea ice cover in the Nordic seas. The presence of a northerly or westerly wind is likely to further promote the formation of a full sea ice cover as it would move the sea ice into the warm boundary current, thereby enhancing sea ice melt, surface freshening, and farther sea ice expansion. Winds would also enhance the turbulent mixing of the ocean, especially over the open ocean. However, we performed sensitivity experiments with increased vertical diffusion, which show little effect on the results, suggesting that the potential increase in vertical mixing with winds would have limited effect on these experiments.
Further studies will focus on the unforced oscillations of the 4DEG experiment. We expect the time scale of the oscillations to change depending on the complexity of the model. In the current model version, the decadal time scale of the unforced oscillations does not compare to the millennial-scale variability observed in glacial climate records. However, the time scales of the unforced oscillations are expected to change when a seasonal cycle is included, as well as feedback with the larger-scale global ocean circulation.
b. Concluding remarks
As noted by Spall (2011), eddy heat fluxes are crucial for a correct representation of the net heat exchange across the sill crest, and thus need to be resolved or parameterized in climate models for a correct representation of the heat transport. We note that the same is true for the advection of sea ice, which is dominated by eddy fluxes at low sea ice cover. Representation of these processes is key to the results presented in this study, which suggests caution should be considered in analyzing similar mechanisms in climate models that are not eddy resolving. In particular, the sensitivity of IPCC-type models to potential future changes in the temperature of inflowing Atlantic water to the Arctic should be analyzed with this in mind.
In this study, a full sea ice cover leads to a reduction in the outgoing heat from the ocean to the atmosphere, a reduction in the overturning streamfunction, and a reduction in the heat transport to the Nordic seas. However, all of the above also decrease with a reduction in buoyancy forcing resulting from decreasing Atlantic water temperatures. As a full sea ice cover develops with a decrease in Atlantic water temperature, the observed changes in circulation could be a result of the decreasing buoyancy forcing. However, the changes are larger with the sea ice transition and show a nonlinear behavior. In addition, Spall (2012) shows periods of reduced convection, overturning circulation, and northward heat transport in similar experiments where the Atlantic water temperature is kept constant. The circulation changes are due to external freshwater input and the establishment of a freshwater cap in the interior that shuts down convection. These points support our conclusion that expansion of sea ice in the Nordic seas is a key player in the circulation changes.
Reconstructions from marine sediments suggest that there were significant changes in the circulation of the Nordic seas and North Atlantic during the stadial to interstadial transitions of the last glacial (Rasmussen et al. 1996; Kissel et al. 1999; Dokken and Jansen 1999; Henry et al. 2016; Burckel et al. 2016). Based solely on the proxy data, it is not clear whether changes in sea ice is a consequence or a cause of these reconstructed changes. However, our study using theory and dynamical models shows that abrupt transitions in Nordic seas sea ice can occur spontaneously with only small, or no, changes in forcing given the appropriate background climate.
5. Summary of key conclusions
The key conclusions are as follows:
There is a strong nonlinear response in Nordic seas sea ice cover to Atlantic water temperatures, with the potential for large changes in sea ice given only small, or no, changes in temperature.
A halocline and Arctic-like stratification emerge through sea ice production given low Atlantic water temperatures.
The advection of sea ice by the mean flow and eddy fluxes is crucial for the development of the halocline.
The presence of a freshwater source is not necessary to explain a fresh surface layer during glacial climates.
Total heat transport to the Nordic seas and the local overturning circulation in the basin decrease with a full sea ice cover.
Eddy heat fluxes dominate the heat exchange across the Greenland–Scotland Ridge.
Given a full sea ice cover, warm inflowing Atlantic water retains most of its heat and recirculates as a warm western boundary current out of the Nordic seas.
The western Nordic seas warms above 800 m and cools below given a full sea ice cover.
The research was supported by the Centre for Climate Dynamics at the Bjerknes Centre for Climate Research. The research leading to these results is part of the ice2ice project funded by the European Research Council under the European Community Seventh Framework Programme (FP7/2007-2013)/ERC Grant Agreement 610055. The simulations were performed on resources provided by UNINETT Sigma2—the National Infrastructure for High Performance Computing and Data Storage in Norway. MAS was supported by the National Science Foundation Grants OCE-1533170 and OCE-1558742. We wish to thank Johan Nilsson and Andreas Born for insightful comments, and the anonymous reviewers for their constructive comments which improved the manuscript.