The mean surface temperature of the earth is rising and how this will influence the circulation of the atmosphere and in the oceans is largely unknown. The observed temperature increase is strongest in the high northern latitudes, which are primarily covered by ocean and one question then becomes: Does the presence of an ocean accelerate or mitigate the observed changes? To address this and related questions we review in Sect. 2 the global radiation balance and how the high north responds to the radiative and mechanical forcing generated by the sun and by the atmosphere. The focus is on the Arctic Mediterranean Sea, which comprises the Arctic Ocean and the Nordic Seas. We also describe the physical and geographical settings and the main bathymetric features of the Arctic Mediterranean and its connections to and communications with lower latitudes in the ocean and in the atmosphere.

Section 3 describes how our knowledge of the Arctic has evolved during the 500 years that have elapsed since the exploration of the high northern latitudes started. This section also reviews how the knowledge of the circulation and the water mass (trans)formations taking place in the Arctic Mediterranean gradually evolved during the about 150 years that oceanographic observations and studies have been conducted in the Arctic Mediterranean. It also discusses the early views of the deep circulation in the world ocean and finally brings up the establishing of the Committee for the Exploration of the Sea (ICES) in 1899 and discusses its importance for the improvement of measurement techniques and theoretical understanding of the ocean circulation.

Section 4 examines the different inflows to the Arctic Mediterranean, the Pacific inflow through Bering Strait and the North Atlantic circulation from the subtropical gyre to the subpolar gyre and into the Nordic Seas. The next sections concentrate on specific features, areas and processes that are presently active and may become more (or less) important in a future warmer climate. Section 5 focusses on sea ice, its physical properties and how its growth and melting influence the underlying water column. Section 6 describes the two inflows from the Nordic Seas to the Arctic Ocean, through Fram Strait and over the Barents Sea, and the interactions (freezing and melting) between Atlantic water and sea ice. Section 7 concentrates on the transformations of the Atlantic water as it moves eastward along the two inflow paths and on the subsequent meeting and remerging of the two inflow branches at the continental slope north of the Kara Sea, Severnaya Zemlya and the Laptev Sea. In this section a wind driven model proposed by Nøst and Isachsen (2003) is also presented.

Eastward of the Laptev Sea the influence of the Atlantic water diminishes and the circulation becomes dominated by freshwater input, both as continental runoff from the Eurasian rivers and as direct inflow of low salinity Pacific water through the shallow Bering Strait (Sect. 8). The upper ocean circulation and mixing are determined by wind and by the seasonal changes in thermohaline forcing, while the deeper (Atlantic) layers become isolated and the observed changes appear due to internal mixing processes such as thermohaline intrusions and double-diffusively active interfaces. In the Transpolar Drift the Siberian branch transports runoff and sea ice from the Laptev Sea, and the anticyclonic Beaufort Gyre brings additional low salinity water towards Greenland. The convergence of these two streams leads to exchanges between the two gyres; the cyclonic gyre in the Eurasian Basin and the anticyclonic gyre in the Amerasian Basin. In Sect. 8 the differences between the deep waters in the Amerasian and Eurasian basins and their causes are also described and discussed.

In Sect. 9 the water masses exchanged through Fram Strait are described from two hydrographic sections, one north of the strait and one along the sill, both taken by FS Polarstern in 2004. Different water masses are identified and the transports of volume, freshwater and heat through the strait are determined from geostrophy combined with inverse methods and a discussion about the reasonableness of the obtained mass and freshwater transports is provided.

The Nordic Seas, especially the Greenland Sea, are characterized by weak stability and are areas of deep convection and deep water renewal, and the convection in the Greenland Sea is discussed in Sect. 10. The water column in the Greenland Sea has exhibited large changes during the last 50 years, indicating that the nature of the active convection processes has changed (compare, for example Carmack and Aagaard, 1973 and Brakstad et al., 2019). This is perhaps the area, where the effects of a warming climate on the water column have been the strongest.

The three overflows of dense water across the Greenland–Scotland Ridge, through the Faroe Bank Channel, across the Iceland Faroe Ridge, and through Denmark Strait, are examined in Sect. 11. The focus is on the Denmark Strait overflow, its possible sources, and the entrainment as the overflow plumes descend into the deep North Atlantic.

The outflows in the upper layer that take place through Fram Strait in the East Greenland Current and through the channels in the Canadian Arctic Archipelago to Baffin Bay and continuing through Davis Strait to the Labrador Sea are discussed in Sect. 12. As the East Greenland Current passes Cape Farewell it turns north as the West Greenland Current and enters Baffin Bay from the south. The transports are modelled as geostrophic flows through control sections with the coast at the righthand side (Werenskiold, 1935). The Canadian Arctic Archipelago and Baffin Bay are considered as a connected and interacting system, and the transports determined from geostrophy are combined with estimates of the surface heat loss and ice formation that determine the upper layer characteristics in Baffin Bay. Estimates of the freshwater storage in the Arctic Ocean are also given and the effects of increased freshwater input from the Greenland icecap are discussed.

The exchanges between the Arctic Mediterranean and the subpolar gyre and the Labrador and Irminger seas are described in Sect. 13, where also the impacts of water mass transformations taking place in a semi-isolated sea may have on the overturning circulation are examined. The final Sect. 14 discusses the global connections and interactions and how they influence and are influenced by the Arctic Mediterranean Sea.

The present review intends to give a comprehensive picture of the Arctic Mediterranean Sea, describing the processes and the water mass transformations in the different parts of the area and how these parts are connected by the circulation of the different waters. The narrative follows two tracks. The first is historical and describes how our knowledge of the geography, the hydrography and the ocean circulation has evolved through time. This approach continues to about 1980 and encompasses the knowledge provided by early summaries of the circulation in the Nordic Seas (Helland-Hansen and Nansen, 1909), the connections between the North Atlantic and the Nordic Seas (Helland-Hansen, 1912; Nansen, 1912, 1913) and the early knowledge of the water masses and circulation in the Arctic Ocean (Nansen, 1902, 1906; Timofeyev, 1960; Coachman and Aagaard, 1974).

After this focus shifts, and the narrative follows the Atlantic water as it enters the Arctic Ocean through Fram Strait and over the Barents Sea, and how it becomes transformed by the processes within the Arctic Ocean, especially by interactions with sea ice and with the freshwater input from the continents and from the Pacific inflow. The Atlantic water is transformed into both less dense water (outflow) and denser water (overflow) within the Arctic Ocean and additional overflow water is formed in the Greenland Sea. The transformed waters eventually return to the subpolar gyre and the North Atlantic.

The review is primarily based on observations and their interpretations. The figures shown thus display mainly temperature and salinity profiles and TS diagrams and vertical temperature and salinity sections. This to show what the interpretations and discussions are based upon. The only schematics are presented in the last figure (Fig. 24).

There are many reviews addressing the Arctic Mediterranean Sea, the Nordic sea and the Artic Ocean. Coachman and Aagaard (1974) give a foundational synthesis of Arctic and subarctic hydrography, circulation, exchange passages, and sea-ice conditions based on the observational knowledge available in the early 1970s. It establishes the Arctic Ocean as a strongly stratified, relatively isolated basin connected to the Pacific and Atlantic through a few narrow gateways. It describes the importance of Bering Strait inflow, the Atlantic inflow via the Nordic Seas and Fram Strait, and the freshwater and ice export to the North Atlantic.

An early modern synthesis of the physical oceanography of the Nordic (or GIN) sea is provided by Hopkins (1991). It is the major physical-oceanography review of the Greenland, Iceland, and Norwegian Seas for the 1970s to the mid-1980s. It describes the basin hydrography, fronts, gyre circulation, the Atlantic Water inflow, the Arctic-origin outflow, deep convection, and sea-ice distribution and explains how the Nordic Seas transform northward-flowing Atlantic waters through cooling, mixing, and densification.

A summary of the pre-IPY (International Polar Year) 2007–2009 knowledge of Arctic water masses and pathways is given in Rudels (2015). It explains the layered structure: the polar mixed layer, the halocline, the Atlantic Water, and deep and bottom waters, and highlights dense-water formation on shelves, boundary current circulation and exchanges with the Nordic Seas and Pacific.

The reviews most akin to the present one are written by Rudels (2012, 2019), Timmermans and Marshall (2020), and Rudels and Carmack (2022). Rudels (2019) gives a concentrated description of the Arctic Ocean circulation and water mass transformations, while Rudels (2012) covers more of the history of Arctic research. Timmermans and Marshall (2020) describe the dynamically distinct Arctic circulation: wind forcing, buoyancy forcing, topographic steering, eddies, and boundary currents. They emphasize how freshwater storage, sea-ice loss, and changing Atlantic or Pacific inflows may reshape stratification and circulation. Rudels and Carmack (2022) discuss the Atlantic and Pacific inflow branches, halocline formation, shelf contributions, and deep-water renewal, and stresses the controlling role of ridges, shelves, and narrow gateways in partitioning water masses among the Arctic basins.

Bluhm et al. (2015) compare the Eurasian and Amerasian basins, including their water-mass structures, circulation, ventilation, and connectivity and show how the warmer, saltier Atlantic influence is more direct in the Eurasian Basin, while the Canada Basin is more strongly influenced by freshwater accumulation and Pacific-origin waters. The review by Wassmann et al. (2020) and book edited by Wassmann (2024) also connects basin-scale physical contrasts to ecosystems and biogeochemistry.

Bacon et al. (2022) concentrates on the Arctic Ocean boundary exchanges and describes the observational estimates of volume, heat, freshwater, and property exchanges through the Arctic gateways. It covers Fram Strait and the Barents Sea Opening in particular, making it directly relevant to the Arctic ocean–Nordic Seas exchange. It emphasizes that gateway fluxes, surface forcing, and storage changes together control Arctic water-mass transformation.

We have perhaps treated sea ice processes and the observed changes in the ice cover rather cavalierly, and we here refer the reader to Meier et al. (2014), which is a broad review of observed losses in Arctic sea-ice extent, thickness, age, and multiyear-ice coverage. It attributes the shift to coupled atmospheric, oceanic, and ice–albedo feedbacks, with summer melt increasingly sensitive to preconditioning. It also discusses implications for Arctic communities, ecosystems, shipping, resource activity, and weather and climate connections

The book by Rudels (2021) contains much of the topics discussed in our review, but it is a textbook, not a review, and contains more detailed descriptions of the circulation and the active processes than are necessary to include in a review article. The book by Mills (2009) provides a very readable account of the history of the ocean sciences.

This review is, admittedly, focused on Western science, as based on the scientific method, and gives little recognition to indigenous contributions to exploration and ecological knowledge. There are two main reasons for this. First, it is beyond the scope of this review, and the authors lack the necessary expertise to do this complex subject justice. Second, the focus of this review is on the circulation and transformations of waters in the deep and offshore Arctic Mediterranean, and thus remote from the Arctic coastal regions where indigenous peoples travel, hunt and live. Regardless, there is much to learn from indigenous thinking as our global society addresses an uncertain future (cf. Berkes and Jolly, 2001; Ford et al., 2010; Falardeau and Bennett, 2019; Falardeau et al., 2019; Huntington, 2011; Huntington et al., 2021). Indigenous knowledge resides in the holder's authority and is experiential and holistic and place-based (Gearheard et al., 2006; Whiteman and Cooper, 2011; Fienup-Riodan and Carmack, 2011). Indigenous knowledge is not a single, uniform body of information, but is learned, practiced, and transmitted differently among distinct Arctic Indigenous Peoples; for reviews, see Eerkes-Medrano and Huntington (2021). Despite the above distinctions, there are key areas where complementarity exists (cf. Carmack and Macdonald, 2008). Intimate connections and dependence of snow and ice conditions can provide an early-warning system for climate and ecological change; one clear example is in documenting local observations based intergenerational knowledge and travel/hunting practice as records of environmental change (Krupnik and Jolly, 2002; Carmack, 2021). That the two-ways of knowing are jointly beneficial was demonstrated by comparison of remote sensing of sea ice and Inuit observations (Meier et al., 2006). Fienup-Riodan and Carmack showed the complementarity of broad spatial records (wind-generated ocean swell) and fine-scale knowledge of ice quality and currents (safety of travel over ice in coastal locations). We emphasize that the depth, breath and sophistication of Indigenous Knowledge of what is Arctic equals that of Western science, and future avenues based on appropriate co-generation of knowledge offer a rich field of inquiry.

The earth receives its energy as shortwave radiation from the sun and its rapid rotation limits the diurnal temperature differences between the sunlit and the dark hemispheres. Large temperature differences arise, however, because of the spherical shape of the earth. The zenith angle of the sun increases with latitude and higher latitudes receive less solar radiation per unit area. The energy received on the luminated hemisphere is radiated back to space as longwave radiation from the entire earth surface. To zero order the energy input and output are globally but not locally balanced and the net heat input at lower latitudes is transported by the atmosphere and the oceans to higher latitudes, where it is radiated back to space. The presence of an atmosphere that contains greenhouse gases that absorb and reradiate longwave radiation leads to a higher mean temperature of the earth than if no atmosphere were present.

The earth's rotation axis makes an angle of 23.5° to the ecliptic. This leads to large seasonal variations in incoming solar radiation, especially north and south of the polar circles (> 66.5°). In summer the incoming daily radiation at the top of the atmosphere is larger at the poles than at the equator (e.g. Fritz, 1951), while in winter the sun is below the horizon and the long, dark polar night reigns. The contrast between the two polar regions is, however, great. In the south the entire ice covered Antarctic continent is in direct contact with the circumpolar Southern Ocean, while in the north there is a confined ocean, the Arctic Mediterranean Sea, with restricted communications to the world ocean and with its northernmost part, the Arctic Ocean, partly covered by perennial sea ice.

The atmospheric poleward heat transport is carried by three distinct circulation cells, the direct Hadley cell where warm air rises at the equator, cools and then sinks back to the surface at 20 to 25° N(S). Some of the sinking air returns towards the equator, forming the trade winds, while the rest is transported poleward by the westerlies in the eddy dominated Ferrell cell. Over the polar areas, in the direct Polar cell, the air is cooled, sinks and returns equatorward, forcing the warmer air of the Ferrel cell to rise.

At low latitudes the poleward oceanic heat transport is larger than the corresponding atmospheric transport, but the transports become equal around 30° N(S) (Wunsch, 2005). At higher latitudes the ocean transport diminishes and beyond 65° only the heat transport in the North Atlantic is different from zero. The Atlantic also differs from the other oceans in that it transports heat across the equator from the southern to the Northern Hemisphere. The warm surface water ultimately comes from the Indian Ocean and is carried by the South Atlantic circulation to and across the equator.

The atmospheric circulation transports not only heat but also water vapor from lower to higher latitudes. In the south this builds up and sustains the glaciers on the Antarctic continent and the freshwater returns as calving icebergs to the Southern Ocean. In the north the water condenses, releasing latent heat and then precipitates, partly over the Arctic Mediterranean, but primarily on the continents surrounding the Arctic Ocean. The drainage area is twice as large as the surface area of the Arctic Ocean and provides a disproportionate freshwater input, 10 % of the entire global continental runoff, to the Arctic Ocean. The created low salinity upper layer leads to strong stability that limits the interaction with the deeper part of the water column.

In summer the net heat input from the sun is larger than the net outgoing longwave radiation and the radiation balance is positive (more heat received than lost). However, the outgoing longwave radiation and the downward reradiated longwave radiation are both individually larger than the downward shortwave solar radiation. During the long winters with no local energy input from the sun the heat advected from lower latitudes cannot balance the radiative heat loss to space and the surface temperature decreases drastically.

The surface air temperature in the Arctic is above the melting temperature of sea water in summer but much below in winter. This leads to phase changes. Sea ice is formed in winter and melts again in summer. The latent heat released by freezing becomes a local heat source that somewhat increases the surface temperature in winter. In summer the returning solar radiation raises the surface temperature and eventually the ice cover starts to melt. The melting ice acts as a thermostat that keeps the surface water temperature close to the melting temperature as long as ice is present. Ice formed and melted does not contribute any net heat to the Arctic Ocean, but the melting in summer stabilizes the water column and isolates the deeper layers from the surface forcing in summer even more than in winter. The ice not melted in summer is eventually exported by the ocean currents southward and becomes a heat source to the Arctic.

The Arctic Mediterranean Sea is a part of the North Atlantic and comprises the Arctic Ocean and the Nordic Seas; the Greenland, Iceland and Norwegian seas, that form an anteroom to the larger Arctic Ocean. The Nordic Seas communicate to the south with the North Atlantic across the Greenland–Scotland Ridge and are connected to the Arctic Ocean through Fram Strait and over the Barents Sea in the north. There is also a direct connection between the Arctic Ocean and the North Atlantic through the narrow straits in the Canadian Arctic Archipelago and then via Baffin Bay and Davis Strait to the Labrador Sea. The Arctic Ocean is connected to the North Pacific through Bering Strait. It is thus not a bay but a convoluted channel between the stratified North Pacific and the convectively ventilated North Atlantic.

The Arctic Ocean is the smallest of the Earth's oceans (9.5 × 10¹² m2), if it could be called an ocean, comprising only about 3 % of the area and less than 1 % of the volume of the global ocean. More than half of the area of the Arctic Ocean consists of broad, shallow shelf seas mainly located north of the Eurasian continent. Farthest to the west is the Barents Sea, separated from the Kara Sea by Novaya Zemlya and Franz Josef Land. The Kara Sea connects eastward to the Laptev Sea through Vilkitsky Strait between Siberia and Severnaya Zemlya. The New Siberian Islands separate the Laptev Sea from the East Siberian Sea, which communicates through Long Sound between the Siberian mainland and Wrangel Island with the Chukchi Sea located north of Bering Strait. Eastward of Alaska the coastline is different. Instead of shallow shelf seas and low lying coasts, it forms the mountainous Canadian Arctic Archipelago, comprising several large islands creating a maze of channels. The most important passages are Lancaster Sound, Jones Sound, and the Nares Strait between Ellesmere Island and Greenland that all connect the Arctic Ocean to Baffin Bay (Fig. 1).

The deeper part of the Arctic Ocean consists of two major basins, the Amerasian Basin and the Eurasian Basin, separated by the narrow Lomonosov Ridge with a mean sill depth around 1600 m. The Amerasian Basin is divided by the Alpha and Mendeleyev ridges into the large Canada Basin and the smaller Makarov Basin, both with maximum depth of the abyssal plains around 4000 m. The Eurasian Basin also consists of two basins, the about 4000 m deep Nansen Basin between the Eurasian continental slope and the Gakkel Ridge and the deeper, 4500 m, Amundsen Basin extending from the Gakkel Ridge to the Lomonosov Ridge. The connection to the Nordic Seas through Fram Strait is deep, sill depth 2600 m. The Greenland Sea comprises two basins, the Greenland Basin with maximum depth about 3500 m and the northerly, smaller and shallower Boreas Basin. The Norwegian Sea is separated into the Norwegian Basin and the Lofoten Basin both about 3000 m deep, while the Iceland Sea is shallower, less than 2000 m deep.

Three deeper passages through the Greenland–Scotland Ridge connect the Nordic Seas to the North Atlantic, the 640 m deep Denmark Strait between Greenland and Iceland, the 400 to 500 m deep ridge between Iceland and the Faroes, and the channel between the Faroes and Scotland, ending in the more than 800 m deep Faroe Bank Channel. West of Greenland the channels from the Arctic Ocean to Baffin Bay are narrow and shallow. Nares Strait has a sill depth of 220 m and Lancaster Sound and Jones Sound have sill depths between 100 and 150 m. Baffin Bay is deep, about 2600 m, and the sill depth in Davis Strait to the Labrador Sea is 630 m. Bering Strait between the North Pacific to the Arctic Ocean is narrow, 85 km, and shallow, sill depth 53 m.

The Arctic Mediterranean Sea is dominated by the interplay between the advection of heat, mass and freshwater from lower latitudes and the local, strongly seasonal, forcing. Most heat is carried by the atmosphere, which also provides the largest freshwater input to primarily the Arctic Ocean, mostly as strongly seasonal runoff from the large catchment areas but also as direct local precipitation (Aagaard and Carmack, 1989; Carmack et al., 2015).

An oceanic inflow to the Arctic Ocean takes place through the shallow and narrow Bering Strait and in the Oceans H. U. Sverdrup estimated this inflow to 0.3 Sv (1 Sv = 10⁶ m6s-1) (Sverdrup et al., 1942). Russian work, however, indicated stronger transports and a significant seasonal variability, 1.2 Sv in summer and 0.4 Sv in winter (Maksimov, 1945; Fedorova and Yankina, 1964). The first direct current measurements made in and north of the strait suggested larger transports, 1.0 to 1.5 Sv, but no seasonal variability was detected (Coachman et al., 1975). The observational work in the strait has continued and later measurements vindicated the early Russian results. The transport varies seasonally with a larger northward transport in summer of 1.2 Sv and a smaller transport, 0.4 Sv, in winter (Coachman and Aagaard, 1981, 1988), and interannually (Itoh et al., 2026).

It has long been known that the northward transport through Bering Strait is due to the higher sea level in the North Pacific than in the Arctic Ocean when adjusted to a baroclinic level of no motion (Coachman and Aagaard, 1966), but Stigebrandt (1984) took a more global view and related the higher sea level in the North Pacific to the less dense Pacific water column, compared to that in the North Atlantic. A difference that ultimately arises from the atmospheric flux of water vapor from the Atlantic to the Pacific Ocean across the Isthmus of Panama (Weyl, 1968). However, also more local processes play a role. The weaker northward transports in winter have been explained by the presence of stronger, more northerly winds (Coachman and Aagaard, 1981, 1988). The inflow across the broad and shallow Bering and Chukchi shelves in summer is warmed, creating the Bering Strait summer water, while the cooling of the inflow in winter forms the Bering Sea winter water, which shows up as a temperature maximum and a deeper temperature minimum, respectively in the Canada Basin water column (Coachman and Barnes, 1961). The Pacific inflow is limited by the shallow sill depth (53 m) of the strait, and thus it does not carry oceanic heat from lower latitudes. Instead, the warmer water transported in summer is formed by local seasonal heating and is a part of the radiation balance of the Arctic Mediterranean. Because of its low salinity the Pacific inflow transports a substantial amount of freshwater to the Arctic Ocean, greatly adding to its stability.

By contrast, the connection between the Arctic Mediterranean and the North Atlantic over the Greenland–Scotland Ridge is deep and wide with a mean sill depth of about 500 m. The thickness of the Atlantic water layer entering the Norwegian Sea is 300–400 m and it transports oceanic heat from lower latitudes to the Arctic Mediterranean. The Atlantic water flows north as the Norwegian Atlantic Current in two branches. One follows the continental slope, while the second branch separates from the slope at the Vøring Plateau and continues as a baroclinic current above the Mohn–Knipovich Ridge towards Fram Strait.

As the Atlantic water at the slope reaches the Barents Sea opening it splits. One branch enters the Barents Sea but then again splits. The major part continues eastward toward Novaya Zemlya, while a smaller stream turns north in the Hopen Deep, where it again splits with one part flowing eastward north the Central Bank and the other returning as cold bottom water to the Norwegian Sea south of Bear Island. The Norwegian Coastal Current that flows between the Atlantic water and the coast, carrying low salinity water from the Baltic Sea and runoff from the Norwegian coast, also enters the Barents Sea and continues eastward close to the coast as the Murman Current.

The branch of Norwegian Atlantic Current that remains at the slope continues northward to Fram Strait, where it partly merges with the outer Mohn–Knipovich branch. Both branches then split. One stream enters the Arctic Ocean along the continental slope close to Svalbard, and a second stream flows around the Yermak Plateau to rejoin the inner branch farther east. A smaller eastward flow crosses the Nansen Ridge that connects the Yermak Plateau to the Svalbard slope into the Sofia Deep (Gascard et al., 1995). A substantial fraction of the Atlantic water, however, recirculates westward and does not enter the Arctic Ocean. Instead, it flows south along the East Greenland shelf and slope, as noticed early by Carl Ryder (Ryder, 1892). The cooling of the Atlantic water in the Norwegian Sea increases its density sufficiently for it to become overflow water that could contribute to the deep return flow of the North Atlantic overturning circulation.

In contrast to Bering Strait the mechanisms driving the Atlantic water into the Nordic Seas are not well known. The warm Atlantic water derives from the Gulf Stream and from the Subtropical gyre. As the warm water in the North Atlantic Drift approaches the British Isles it splits. The upper layer moves mainly south and remains in the subtropical gyre, while the deeper layers, below 300 m, raise to the surface and flow north into the Subpolar gyre (Brambilla and Talley, 2006; Burkholder and Lozier, 2011). The functioning of this valve and how it might influence the properties of the northward flowing Atlantic water is not adequately known but likely depends on a combination of the characteristics of the North Atlantic Drift, the circulation in the Subpolar gyre and on the properties of the waters exiting the Arctic Mediterranean.

The more fundamental question is if the inflow across the Greenland–Scotland Ridge is driven by winds and large weather systems like the North Atlantic oscillation, or if the cooling of the Atlantic water and the convection taking place in the Nordic Seas draw the warm Atlantic water across the ridge in the manner envisaged by Carpenter? This will have a large impact on how the Arctic Mediterranean responds to changes in climate and we return to this question later.

The most outstanding feature of sea ice is that it floats. It is not unique but rare that a substance, when solidified by cooling, becomes less dense than its liquid phase. In freshwater this is preceded by a density maximum at 4 °C, above the freezing point, which means that further cooling creates its own stable stratification, limiting the amount of water that must reach freezing temperature before ice is formed. Sea water with salinity S > 24.7, by contrast, is densest at its freezing point and the water column has to be convectively cooled to freezing temperature down to the bottom, or down to a density gradient caused by the presence of a deeper more saline water mass, before sea ice is formed. As the water column is at the freezing point no sensible heat is supplied by the underlying water and the only heat source for the atmosphere is the release of latent heat of freezing, and a balance between the heat lost at the ice surface and the heat conducted through the ice is created. A balance that gradually shifts as the thickness of the ice cover increases and the ice surface temperature decreases, which reduces the ice growth at the ice-sea interface.

The solid ice cover allows snow to accumulate during winter, which isolates and further reduces the heat loss to the atmosphere and to space. A white snow cover implies a high albedo and when the sun rises in spring most of the shortwave radiation is reflected back to space. This is perhaps the most often mentioned Arctic feedback mechanism. Warmer air, less snow, more absorbed shortwave radiation, warmer air, less snow, etc. However, another mechanism that may be even more important is increased advection of moist air from lower latitudes, which increases the downward longwave radiation and may start melting snow and ice before sunrise (Mortin et al., 2018).

During summer the incoming solar radiation increases the air and water temperatures and snow and sea ice start to melt, creating an upper melt water layer that, however, remains close to melting temperature as long as ice is present. If the ice cover is not compact, shortwave radiation penetrates into and through the melt water layer and creates a warmer layer below. This heat is isolated from the ice until autumn, when ice again starts to form and the salinity of the melt water layer increases. These two layers might then merge, which would temporarily slow down the seasonal ice formation. If not, the warmer layer becomes a permanent temperature maximum in the water column, the Near Surface Temperature Maximum (Jackson et al., 2010, 2011).

A stable seasonal cycle of ice formation and ice melt would eventually lead to an equilibrium thickness of the ice cover. The equilibrium is reached because the ice formation in winter depends more upon the ice thickness than the ice melt in summer. Thinner ice leads to rapid growth in winter and the ice cover becomes thicker. If the ice is initially thick, little ice growth takes place in winter, while a larger amount melts in summer and the ice thickness decreases (Maykut, 1986).

As sea ice forms, either by growing directly from the underside of the ice cover or created as frazil ice in the water column, salt and freshwater become separated into brine and ice. The sea ice becomes thicker and the water below saltier, and in the case of frazil ice formation the ice crystals rise to the surface and “droplets” of brine sink. The smaller brine droplets soon lose their identity as they sink and the salt diffuses into the surrounding water. This creates a thicker, slightly unstable layer that eventually must convect into the underlying ocean (Rudels, 1986a). However, the frazil ice that is collected as a “mushy” layer (Wettlaufer et al., 1997) at the surface traps a large fraction of the brine, and as the mushy layer solidifies the brine is eventually squeezed out of the ice and sinks. The brine parcels rejected out of the mushy layer are likely larger than the droplets formed together with the frazil ice and might sink as individual parcels to greater depth before the salt anomaly is diffused away.

Until now we have only looked at the case when the upper layer temperature is close to the freezing/melting temperature and not considered what happens, when sea ice encounters and melts on warmer water. Nansen (1912) studied this situation by laboratory experiments. A vessel with warm, saline water was cooled from above in three different manners. First a jar with sea ice was held above the water surface and the water was cooled by radiation. Then the jar was allowed to float on the water, cooling the water by conduction. Finally, the ice was added at the surface and coming in direct contact with the ice (Fig. 6). The temperature evolution at the bottom and at the center of the vessel was measured in all experiments. The difference between the three experiments is striking. In the case of just radiative cooling the temperature decrease of the water is small and slow, while the conductive cooling of the water leads to a rapid and large temperature change in the vessel, indicating that thermal convection is established. The experiment where the ice is melting directly at the sea surface is the most revealing. A short rapid cooling takes place but after that the temperature remains essentially constant. The warm water melts the ice, which leads to the formation of a cold, less saline and less dense surface layer that isolates the ice from the warm water below. Sea ice evidently protects itself from heat input from below by creating a stratification that limits the heat transfer.

Nansen's experiments were idealized. No turbulence was present in the vessel and the experiments were performed at room temperature. In the ocean wind will generate turbulence, either directly or by driving the ice, which entrains water across the interface into the upper layer. Furthermore, in summer the main sea ice melting occurs at the surface and the melt water layer prevents heat from below to reach the ice. In winter, however, the air temperature is low and the summer melt water has again become ice. As the warm water gets in contact with and melts the ice, an upper, cooler and less saline and less dense layer develops and deepens as warm water is entrained from below.

Assuming that the deepening of the upper layer can be described by a Kraus-Turner energy balance model (Kraus and Turner, 1967; Niiler and Kraus, 1977), where the turbulent energy added at the surface is used not only to entrain denser water from below but also to mix melt water from above into the mixed layer the entrainment velocity we becomes: 1we=2nou∗3g[β(S2-S1)-α(T2-T1)]-εBg[[β(S2-S1)-α(T2-T1)]]

Here no is a constant usually put to 1.25, u∗ is the friction velocity, T1, T2, S1 and S2 are the temperatures and salinities of the upper and lower layer, respectively, α is the coefficient of heat expansion and β the coefficient of salt contraction. The sign of α is explicit in the expression. g is the acceleration of gravity, B is the buoyancy input and ϵ is a parameter put to 1 if the buoyancy input is positive and 0.05 if the buoyancy is negative to take the weak mixing rate of convection into account (Stigebrandt, 1981). If the density is taken only to depend upon salinity, the second term becomes equal to we and the input of fresh (melt) water at the surface thus reduces the entrainment by a factor of 2 (Rudels, 1989).

The entrained heat not only melts ice, in winter the heat loss to the atmosphere also has to be supplied by the entrained water. The distribution of the entrained heat to sea ice melting and to the atmosphere is crucial for the evolution of the upper layer, and one suggestion is that the fraction of heat going to ice melt is the one that leads to minimum ice melt (Rudels et al., 1999a) This fraction ϕo is found to be: 2ϕo=2αLc(βS2-αΔT)≈2αLcβS2

L is the latent heat of melting and c the heat capacity of sea water and ΔT the temperature difference between the Atlantic and the melt water layer. (Rudels et al., 1999a). α decreases with decreasing temperature and Fig. 7 shows the total heat loss Q′, the heat loss to ice melt ϕoQ′, and the heat loss to the atmosphere (1-ϕo)Q′ as functions of ϕ for different ΔT and the corresponding values of α.

As the created upper layer deepens its salinity and temperature decrease, while the stability increases with time until the freezing temperature is reached. The salinity S1 in the upper layer is then given by: 3S1≈S21+2αβS2(T2-Tf)

T2 is the temperature of the lower layer and Tf the freezing temperature of the upper layer. The stable salinity step β(S2-S1) is about twice the unstable temperature step α(T2-T1) giving a stability ratio of 2 at the interface. The problem is the choice of α, which depends strongly on temperature and as the upper layer deepens and cools α decreases to a minimum at the freezing point. A smaller α leads to a deeper upper layer with higher salinity.

When the freezing temperature is reached, the upper layer has become so deep that the amount of entrained warm water is not enough to supply heat both to ice melting and to the atmosphere, and ice again starts to form. The entrained oceanic heat loss, however, is still expected to be distributed between ice melting and the atmosphere that the ice melting is a minimum. The stability at the interface now decreases due to ice formation and brine rejection and the upper layer eventually becomes dense enough to convect into the deep underlying water (Rudels et al., 1999a; Rudels, 2021). This situation will be discussed in more detail in Sect. 10 below. Here we might just add that the fraction ϕo of minimum heat going to ice melt is twice the fraction leading to marginal stability. A fraction smaller than ϕo/2 will not stabilize the water column and it overturns. This is also evident from the stability ratio Rρ = βΔS/αΔT = 2 reached when the upper layer has attained freezing temperature.

The Atlantic water that enters the Arctic Ocean through Fram Strait soon encounters and starts to interact with sea ice. In winter the melting sea ice and entrained Atlantic water create a less saline melt water layer that is eventually cooled to freezing temperature (Rudels et al., 1996). As a result, an area of open water often forms north of Svalbard, the “Whalers' Bay”. Whether heat entrained by mechanical mixing from the Atlantic layer below is sufficient to melt the ice and create a polynya, or if convection of the entire upper layer, driven by ice formation and brine rejection after it has reached freezing temperature, is also required, remains an open question. The Whalers' Bay normally becomes ice covered later in winter, either by locally formed ice or by ice advected from the interior of the Nansen Basin.

The formation of the less saline upper layer implies that part of the Atlantic water that was cooled in the Norwegian Sea and transformed into high density water, which potentially could contribute to the overturning circulation, instead will leave the Arctic Mediterranean as a less dense upper outflow. The now stratified Atlantic water column continues as a boundary current eastward along the Eurasian continental slope and in the following summer sea ice melting by shortwave radiation creates a seasonal surface melt water layer.

The fate of the water that enters the Barents Sea is more varied and more difficult to follow. The Norwegian Coastal Current and the main Atlantic inflow branch flow eastward towards Novaya Zemlya and the heat advected by the Atlantic water is sufficient to keep the southern half of the Barents Sea free of ice the year around. The heat loss here is large and the Atlantic water becomes considerably colder than the Atlantic water entering the Arctic Ocean through Fram Strait. In the southern Barents Sea ice is only formed close to the coast and over the shallow areas west of Novaya Zemlya, which are mainly covered by water from the less saline coastal current and where the entire water column is cooled to freezing in winter.

The northern Barents Sea is different. It is shallow, mainly comprising the about 120 m deep Grand Bank. It connects to the Arctic Ocean between Svalbard and Franz Josef Land with the deepest trough being the 500 m deep Victoria Channel west of Franz Josef Land. The northern Barents Sea is strongly stratified and seasonally ice covered. The source(s) of the low salinity upper layer is(are) not obvious. Its characteristics suggest that it is not directly supplied from the Arctic Ocean, but it might enter from the Kara Sea in the Persey Current, which flows westward south of Franz Josef Land. Another possibility is that the upper layer is formed in the Barents Sea by sea ice melting on the warmer Atlantic and coastal waters that move north towards the passage to the Kara Sea between Franz Josef Land and Novaya Zemlya. The low salinity, cold upper water is eventually exported from the northern Barents Sea to the Norwegian Sea in the Bear Island Current. The more saline and denser deep and bottom waters in the northern Barents Sea are supplied by the branch of Atlantic water that mainly recirculates in the Hopen Deep, but partly penetrates northward onto the Grand Bank.

In the 2010s the ice cover in the northern Barents Sea was significantly reduced and the stability between the upper layer and the Atlantic water below weakened. This was suggested due to an increased inflow of warmer Atlantic water that reduced the stability and melted the sea ice from below (Lind and Ingvaldsen, 2012; Lind et al., 2018). The change would be part of a larger shift that is presently occurring in the Arctic Ocean, the so called “Atlantification” of the Arctic Ocean, which implies a reduction of the upper layer thickness and stability, increased shear and a stronger upward transport of heat from the Atlantic layer to the ice (Årthun et al., 2012; Polyakov et al., 2017; Årthun et al., 2025). This phenomenon was first reported from the eastern Nansen Basin north of the Laptev Sea, and the changes observed in the Barents Sea were taken as signs that the same pattern was expanding. The heat advected from the North Atlantic would in the near future dominate not only the Norwegian Sea but also the Nansen Basin and the Barents Sea and reduce the extension of the ice cover (Lind et al., 2018; Polyakov et al., 2020a, b). In the later part of the 2010s the reduction of the ice cover in the northern Barents Sea stopped and it started to recover, but it is still far from its extent in the 1980s.

Not only the ice cover but also the water mass characteristics in the Barents Sea have changed considerably during the last decades. In the 1980s the ice cover and the low salinity upper layer extended farther south and occasionally covered the Central Bank. In some years ice formation and brine rejection over the bank led to haline convection to the bottom and the formation of a cold, dense water column over the bank that eventually contributed to dense water formation in and dense water export from the Barents Sea (Quadfasel et al., 1992; Schauer et al., 2002). In years with strong inflow of warm Atlantic water the Central Bank, by contrast, became covered by Atlantic water and no ice and no dense water was formed over the bank. In the 2000s and 2010s the Atlantic water has been warmer, more saline and denser than in the 1980s. The convection over the Central bank has been thermal and the created waters have been denser than those formed by brine rejection in the 1980s.

The dense water formation in the Barents Sea has altered not only over the Central Bank but also farther east, above the Central Depression and over the shallow areas west of Novaya Zemlya. The Atlantic water cooled in winter is presently denser than in the 1980s and shows up as warmer bottom water below a less saline water mass with temperature close to freezing (Lien and Trofimov, 2013; Dmitrenko et al., 2015). The salinity indicates that the upper layer could be created by sea ice melting on top of the denser Atlantic water (Eq. 3 above). This is in contrast to observations made in the early 1900s, when the water created by ice formation and brine rejection west of Novaya Zemlya was the densest water formed on the Barents Sea (Knipovich, 1905; Nansen, 1906). Dense water formation by brine rejection still occurs but now it appears to mainly take place north rather than west of Novaya Zemlya (Lien and Trofimov, 2013).

The downstream conditions of the inflow to the Arctic Ocean though Fram Strait and the upstream conditions of the return flows to the strait are shown in the northern section in Fig. 11. The inflow involves not only Atlantic water, which mainly enters over the shelf north of Svalbard, but also denser, deeper lying waters, which cannot pass directly eastward and have to circulate along the Nansen Ridge and the northern slope of the Yermak Plateau before they can join the shelf inflow at the continental slope east of Svalbard. Farther north waters returning towards Fram Strait in the northern Nansen Basin, over the Gakkel Ridge and in the Amundsen Basin dominate. Both the Greenland Sea Deep Waters and the Eurasian Basin Deep Waters above the sill of Fram Strait have warmed in recent years, such that the Greenland Sea is now a source of warming to the interior Arctic Ocean (Karam et al., 2024; Song et al., 2025).

The Arctic Atlantic water returning in the Eurasian Basin is warmer and more saline than the Arctic Atlantic water that has circulated through the Amerasian Basin and approaches Fram Strait along the Greenland slope. The water that enters the Arctic Ocean over the Barents Sea is identified as a colder, less saline layer at 1000 m depth, but it is difficult to separate the intermediate waters recirculating in the Eurasian Basin from those that have passed through the Amerasian Basin, except by location. The deep water from the Amerasian Basin shows up as a salinity maximum around 2000 m at the Greenland slope, and the colder and denser Eurasian Basin deep water fills the deeper layers on the northern section (Fig. 11).

On the southern section, taken along 78° 50^′ N in Fram Strait (Fig. 11), there is an inflow of mainly Atlantic water in the West Spitsbergen Current but also of intermediate and deep waters from the Nordic Seas. The deep waters in general are slightly less saline in Fram Strait than farther north in the Nansen Basin. The central part of the strait exhibits a strong recirculation of Atlantic water towards the west. The recirculation is largely carried by barotropic and baroclinic eddies that appear to move westward towards Greenland. The main southward flow from the Arctic Ocean takes place in the west in the East Greenland Current. The recirculation in Fram Strait is strong and only a narrow space is available for the outflowing water between the recirculating Atlantic water and the Greenland continental slope.

The East Greenland Current is more baroclinic than the West Spitsbergen Current and carries several different water masses. The upper layer is dominated by low salinity waters comprising river runoff, occasional contributions from the Pacific inflow, and halocline water created from ice melting on the warm Atlantic water entering north of Svalbard. The cold, low salinity water flows above warm recirculating Atlantic water that penetrates into the East Greenland Current above the transformed Arctic Atlantic water that is returning along the Greenland continental slope. The warmest Arctic Atlantic water derives from the Fram Strait branch that mainly remains in the Eurasian Basin while the colder contributions are supplied by the Barents Sea inflow branch that circulates around the Makarov and Canada basins. The Barents Sea inflow is transformed into an intermediate water mass with distinct doubly stable θS characteristics, salinity increasing and temperature decreasing with depth.

At the western slope two outflows of the Arctic Ocean deep waters can be identified. A warm, saline core centered around 1800 m depth, which ultimately derives from the Amerasian Basin, and a colder core with similar salinity located at sill depth, indicating an outflow of Eurasian Basin deep water. The comparatively warm and saline characteristics of the Arctic Ocean deep waters indicate that they are created by slope convection as shelf water made dense by ice formation and brine rejection sinks down the slope entraining ambient warmer water. The salinity then decreases and the temperature increases and the plumes continue to sink until their density matches that of the surrounding water column. The deep and bottom water of the Arctic Ocean thus become warmer and more saline than deep water created by open ocean convection. The entrainment into the boundary plumes transports Atlantic water downward in the water column and its density increases as it mixes with the cold, saline shelf outflow. The interaction with sea ice formation and sea ice melting thus separates the inflowing Atlantic water into both less dense and denser waters.

Fram Strait is the only deep passage to the Arctic Ocean and most of the entering Atlantic water must exit through Fram Strait. The low salinity waters can, however, leave the Arctic Ocean either through Fram Strait or through the narrow and shallow passages of the Canadian Arctic Archipelago. The largest runoff enters the Kara and Laptev sea shelves and since the shelf outflow from the Laptev Sea to the Amundsen Basin has salinity between 32 and 33 the runoff must mix with an advective saline water mass, and the only one available on these shelves is the water of the Norwegian Coastal current that enters the Barents Sea and continues eastward to the Kara Sea and Laptev Sea shelves.

The Chukchi shelf is flooded from behind by the Pacific inflow, which also supplies saline water to the East Siberian Sea, where, in addition, upwelling of saline water from the Makarov Basin takes place. The melting of sea ice on the entering Atlantic water, either north of Svalbard or in the Barents Sea, creates the densest upper layer water and the densest water that can pass through Nares Strait, the deepest passage in the Canadian Arctic Archipelago (Rudels et al., 2004; Dickson et al., 2007).

The major driving forces behind the exchanges through Fram Strait are not identified with certainty, but the largest transports take place at the slopes, which could indicate either a barotropic wind driven flow along geostrophic contours as suggested by Nøst and Isachsen (2003), but also baroclinic transports of mainly less dense upper layer waters along the coastal boundaries (Werenskiold, 1935; Stigebrandt, 1981), or a combination of both. The sea level in the Arctic Ocean is higher than in the Nordic Seas, which favors a southward flow in the East Greenland Current. It could also obstruct the northward flow in the West Spitsbergen Current and force the westward recirculation in the strait.

The presence of eddies in Fram Strait leads to weak horizontal coherence of the velocities and makes it difficult to determine the transports by direct current measurements. Existing hydrographic sections, by contrast, have a much higher horizontal and vertical resolution which allows for better identification of the transports of different waters. The temporal resolution is of course much less, one section per year compared to one record per hour. Several attempts to compute the geostrophic volume transports through the strait and to estimate the heat and freshwater fluxes have been made. One weakness always present is the choice of the unknown barotropic reference velocities, which to some extent must be subjective. However, this and other difficulties may also offer opportunities to reflect upon what the obtained results really imply and we will discuss in some detail the exercise made by Rudels et al. (2008).

Rudels et al. (2008) noticed that the deep water from the Arctic Ocean is more saline than the deep inflow from the Nordic seas, and that the deep inflow also has to leave the Arctic Ocean as deep water (28.06 ≤ ρθ). They postulated that 0.4 Sv of deep water (S = 34.935) is produced in the Arctic Ocean and that 0.2 Sv of deep water (S = 34.910) enter from the Nordic Seas. This implies a net deep outflow M, of 0.4 × 10⁶ m3s-1, carrying a net salt export S of 13.973 × 10⁹ kgs-1. They applied these constraints on the deep part of the strait and required that the kinetic energy of the added barotropic reference velocities should be a minimum. The barotropic reference velocities extend to the sea surface, but no corrections are applied on the shallow parts of the strait, where no deep water is present. The applied constraints promote a saline deep outflow on the western and the less saline inflow on the eastern side of the strait, in accordance with observations.

Rudels et al. (2008) introduced six different water masses and computed the in- and outflow of these waters in four different parts of the strait; the Greenland shelf and slope, the western and the eastern central parts of the strait, and the Svalbard shelf and slope for each section. The greatest differences were seen between the Greenland and the Svalbard shelves and slopes, while the water flowing north and south in the central part of the strait were similar, indicating a strong recirculation.

The northward and southward volume transports in Fram Strait are not balanced. This implies that the computed heat and freshwater transports depend upon the choice of reference temperature and reference salinity. Rudels et al. (2008) put the reference salinity equal to the mean inflow salinity and the reference temperature equal to the mean outflow temperature. This implies that the inflow carries no freshwater and the outflow no heat.

Computing the geostrophic transport through 16 hydrographic sections obtained between 1980 and 2005 along the sill in Fram Strait the total in and outflows ranged from 5 Sv to almost 15 Sv with an averaged inflow of 6 Sv and an average outflow close to 9 Sv. This is, as expected, less than the transports obtained by direct current measurements, but not alarmingly so. The mean net outflow, 2.5 Sv, 2.8 Sv if the southward flow on the Greenland shelf is included, is larger than that found from the current meter array.

The transports through the other passages of the Arctic ocean are only in one direction. Inflows take place over the Barents Sea and through Bering Strait and runoff and net precipitation add freshwater. The only other outflows occur through the Canadian Arctic Archipelago. The shelf input and the Pacific inflow are cooled to freezing temperature but the part of the Barents Sea inflow that enters the deeper layers of the basins has mean temperatures around -0.5 °C. We assume that the transports through the other passages are known and use the estimates provided by Dickson et al. (2007). It is then possible to determine the amount of heat entering and the amount of freshwater leaving the Arctic Ocean through Fram Strait and also how the advective heat is lost in the Arctic Ocean.

The net volume transport, the mean outflow temperature and the northward heat transport referenced to the outflow temperature were computed for each section. The average values 2.8 Sv, Tref = 0.7 °C and 25 TW obtained from the sections were taken to represent the transports and used to determined how the inflowing heat might be distributed in the Arctic Ocean.

Assuming that 1.2 Sv of the Barents Sea inflow with temperature -0.5 °C enter the deeper layers and leave the Arctic Ocean through Fram Strait, this cold water transports c×(Tref-(-0,5))× 1.2 × 10⁹ = 5.8 TW into the Arctic Ocean. c = 4000 Jkg-1K-1 is the heat capacity of sea water. The remaining part, 1.6 Sv, of the net outflow is less saline and at the freezing point and exits as upper layer water transporting c×(Tref-(-1,8))× 1.6 × 10⁹ = 16 TW heat northward. The northward heat transport in the East Greenland Current of 22 TW then has to be balanced by outflowing warmer returning and recirculating Atlantic water. This implies that almost all 25 TW of the northward heat transport through the strait return south as warmer water in the East Greenland Current.

The remaining 3 TW could melt sea ice north of Svalbard and with an Atlantic inflow temperature of 3.1 °C this implies that 0.15 Sv is cooled to freezing temperature and that the fraction ϕ = 2αL(cβSA)-1 (see Sect. 5) or about one third of this heat loss, is used to melt ice and the rest goes to the atmosphere. The amount of melted ice added to the created upper layer is then 0.003 Sv.

Introducing the obtained freshwater and volume transports through Fram Strait in the volume and freshwater balances for the Arctic Ocean compiled by Dickson et al. (2007) Rudels et al. (2008) found that the freshwater fluxes into and out of the Arctic Ocean were almost in balance but the volume transports were unbalanced by 1.17 Sv. This suggests that the barotropic velocities added in Fram Strait are too small. However, since the chosen reference salinity is the mean salinity of the inflowing water, we can add a barotropic correction to the northward transports without affecting the freshwater balance. If a barotropic northward transport of 1.1 Sv is added also the volume fluxes become almost balanced. The northward heat transport, however, then increases.

The difference between the mean inflow temperature, 1.6 °C, and the mean outflow temperature 0.7 °C is 0.9 °C, which increases the heat transport by 4 TW. The net outflow that has to be heated to 0.7 °C decreases from 2.8 to 1.7 Sv, implying that only 0.5 Sv of the cold surface water must be warmed to 0.7 °C, which requires 5 TW. The heat transport through Fram Strait that enters the Arctic Ocean then becomes 19 TW. If all this heat goes to ice melt by cooling the Atlantic water to the freezing point about 1 Sv of Atlantic water is converted to low salinity upper water. 7 Tw go to melting sea ice and 12 TW are lost to the atmosphere. This means that 0.021 Sv of ice is melted and the salinity of the upper layer becomes 34.3.

These values are realistic and agree with other, more recent estimates of the Arctic Ocean heat and freshwater balances (e.g., Beszczynska-Möller et al., 2011; Rudels et al., 2015; Rudels, 2016; Bacon et al., 2022). It appears that the heat transport and the effects of the heat carried by the Atlantic water into and around the Arctic Ocean are less than commonly assumed. The existence of strong warm inflow pulses of Atlantic water could perhaps alter this situation and increase the effects of the Atlantic inflow. However, the persistence of the high temperatures in the Atlantic inflow pulses as they circulate around the Arctic Ocean basins (Polyakov et al., 2005, 2011) rather indicates that the heat remains in the Atlantic layer and the heat is eventually returned to the North Atlantic through Fram Strait.

The discussion above is largely ad hoc and relies strongly on subjective reasoning. However, the inverse modeling of the exchanges between the Arctic Ocean and the world ocean has greatly improved as more observations from the different gateways have become available. Tsubouchi et al. (2012) estimated the transports through all four passages (Davis Strait, Fram Strait, Barents Sea and Bering Strait) for August 2005 when all gateways were monitored and found an oceanic heat flux into the Arctic Ocean of 154 ± 44 TW and a freshwater export from the Arctic Ocean of 155 ± 65 mSv relative to the mean temperature 1.01 °C ± 0.16 °C and the mean salinity 34.67 ± 0.02 on the total cross section. Since all passages were monitored the total heat and freshwater transports are here independent of reference temperature and salinity. However, the heat and freshwater transports through the individual straits may occasionally appear somewhat odd. The largest freshwater export takes place through Fram Strait but primarily as import of saline Atlantic water in the West Spitsbergen Current, not as export of low salinity water in the East Greenland Current.

To estimate the water transformations taking place in the Arctic Ocean Tsubouchi et al. (2012, 2018), Brown et al. (2025) and Bacon et al. (2022) computed the in- and outflows in different density intervals, and the resulting imbalances had to be balanced by mixing across the density surfaces within the Arctic Ocean. The computations revealed that the density interval encompassing the Atlantic water was not balanced and of the inflowing 6.0 Sv only 3.2 Sv returned in the same density interval as modified Atlantic water. 1.0 Sv had become diluted by freshwater and returned as less dense upper layer water, while 1.8 Sv had been transformed into denser water and returned as intermediate and deep waters (e.g., Bacon et al. 2022). The formation of upper layer water is mainly due to sea ice melting north of Svalbard and the creation of denser water takes place by atmospheric cooling largely in the open water area in the southern Barents Sea.

The observations at the gateways thus reveal transformations of the waters within the Arctic Ocean, processes that are not considered in the circulation model by Nøst and Isachsen (2003). The results above, however, are based mainly on observations, but Dörr et al. (2026) showed that the output from a global NEMO ocean hindcast simulation can describe many of the active processes and how they transform the waters entering the Arctic Ocean. The model was forced with atmospheric and runoff data from the DRAKKAR forcing set and was run from 1958 to 2015 and the outputs from 1979 to 2015 were examined. The computed net in- and outflows through the different passages are close to those observed, there are net outflows through Davis Strait (1.8 Sv) and through Fram Strait (3.0 Sv) and inflows over the Barents Sea (3.5 Sv) and through Bering Strait (1.3 Sv).

The focus is on water mass transformations and the circulation is described both in density space, showing the transformation of inflowing water into denser and less dense waters, similar to the approach used by Brown et al. (2025). However, the transformations are also shown in TS space, following the approach introduced by Gösta Walin (1982). The model results in similar water mass transformations as those deduced from observations in the Arctic Ocean and discussed in the sections above. One significant difference, however, is that about 3 Sv of Atlantic water are transformed into denser water (Dörr et al., 2026), almost twice the transformation estimated by Brown et al. (2025) and Bacon et al. (2022). The main area of this transformation was also here located to the southern Barents Sea.

The Nordic Seas were early identified as a possible source area for the cold, dense deep waters of the oceans. Wyville-Thomson and Carpenter found cold, dense bottom water south of the Faroes, Mohn identified a cold, almost homogenous water column southeast of Jan Mayen, and Amundsen's observations from Gjøa in the Greenland Sea in 1901 showed a cold homogenous dome of water extending from the bottom almost reaching the surface (Nansen, 1906). However, Nansen (1912) judged that the deep convection observed in the Greenland Sea had little global impact and instead postulated that the bulk of the ocean deep water is formed farther south, in the Irminger Sea southeast of Greenland, or in the Labrador Sea.

In the 1970s this view began to change. Satellite observations showed a polynya in the Weddell Sea ice cover 1977 to 1979 and hydrographic observations confirmed that the polynya was related to a cold, less saline water column indicating recent convection (Gordon, 1978). This stimulated a search for and studies of convection, or chimneys (Killworth, 1979), and in the Greenland Sea satellite observations showed several rapid events of ice formation and subsequent disappearance of sea ice, suggesting sinking of cold brine enriched surface water, which was replaced by upwelled warmer water, melting the ice.

Convection occurs as small-scale short events in winter, usually during bad weather conditions making it difficult to observe and study. In the 1970s field experiments dedicated to convection studies were initiated in known convection sites, in the Mediterranean Sea and in the Labrador Sea, and in the 1980s and 1990s most observational and theoretical work in the Greenland Sea was concentrated on convection and many new techniques were introduced. At the same time the importance of the Greenland Sea convection for the formation of dense deep water that could drive the Atlantic meridional overturning circulation, or the global thermohaline circulation as it was then commonly called, was realized, and perhaps more urgent, what were the consequences, if the convection and deep water formation should weaken. In retrospect, in 2026, the three decades between 1980 and 2010 were decades when the largest known changes in the Greenland Sea convection occurred, and when our understanding of the deep water formation in the Arctic Mediterranean deepened.

In the mid-20th century the principal sign of deep convection, a homogenous water column extending from the surface to the bottom, had not been observed in the Greenland Sea and the early observations by Mohn and Amundsen were not fully trusted because the instruments used then were not of the same quality as those deployed during the 1950s. Metcalf (1955) suggested a sloping descent of the dense water that would not be captured by vertical hydrographic casts but this idea was never really accepted. The water column in the central Greenland Sea had structure that appeared to be conserved, even if its deep and bottom waters were renewed. Generally, it consisted of a low salinity upper layer above a deep temperature maximum and further down a salinity maximum below which the temperature and salinity decreased with depth until the homogenous Greenland Sea bottom water was encountered. How could this structure be explained?

Carmack and Aagaard (1973) suggested that the deep and bottom water formation was driven by double-diffusive transports through diffusive interfaces. Atlantic water circulating around the Greenland Sea gyre penetrates towards the center while losing heat upward to the colder surface water but retaining its high salinity. When it finally reaches the center of the gyre, its density has become high enough to renew the deep water and perhaps sink to the bottom. The fact that the convecting water is not in direct contact with the atmosphere could also explain the low oxygen content observed in the Greenland Sea deep water (Carmack and Aagaard, 1973). The importance of double-diffusion was further explored by McDougall (1983), who found that the amount of dense water that could be created by this process corresponded well with existing estimates of deep water formation in the Greenland Sea.

Deep convection areas are usually related to cyclonic circulation that forces the surface water towards the rim and allows the deeper layers at the center to rise as a dome towards the surface. A smaller volume of upper layer water must then be cooled before convection starts and the water column overturns more readily (Killworth, 1979, 1983; Marshall and Schott, 1999). The smaller volume also allows for the possibility that the convection occurs in a filling box mode, where the upper layer density increases in winter, becomes unstable and convects to the bottom, bypassing the intermediate layers (Baines and Turner, 1969). The difficulty with this explanation, in contrast to e. g. brine rejection over the shelves, is that it is not possible to accumulate dense water at the surface. A Rayleigh number based on the ocean depth would directly make the water column unstable, and even with a Rayleigh number based on the penetration depth of the heat anomaly from the surface (Howard, 1966) no large positive anomaly could be created before convection starts.

This fact and satellite observations showing an interplay between the ice tongue Odden and the ice free Nordbukta, suggesting rapid formation and melting of the ice cover in the central Greenland Sea, opened for the possibility that open ocean deep convection could be driven by ice formation (Rudels, 1986a). As the water is cooled to freezing temperature, ice will form and rise to the surface, leaving more saline water droplets behind. These droplets sink through and their excess salinity diffuses into the ambient upper layer. As the droplets eventually lose their salinity it is instead the ambient upper layer that has become saline and dense with respect to the underlying water and starts to convect. The density and thickness of the unstable layer can be estimated from how long it takes for the salinity anomaly to diffuse, using molecular diffusion coefficients, into the ambient water, and how fast the parcels sink, computed from Stokes' resistance law.

The unstable upper layer then sinks as entraining plumes or thermals into the deep until the plumes reach their neutral density level, where they spread out. Convection, however, continues and the upper part of the water column becomes gradually denser until its density is high enough for the convection to penetrate still deeper. The transformed upper layer then convects into the deep and becomes replaced by warmer water from below that starts to melt the ice and the convection stops. If the cooling persists, the upper layer is again cooled to freezing temperature and ice starts to form, brine is rejected, and another convection cycle starts (Rudels, 1986a: Rudels et al., 1999a). Since it is cold water that convects, it is possible that the thermobaric effect, arising because cold water is more compressible than warm water, further increases the density of the plumes, if the convection reaches sufficiently deep.

This scenario is highly speculative, perhaps even controversial, and field studies were initiated in the Greenland Sea to capture at least some of the involved processes (Quadfasel and Rudels, 1990). During strong cooling and ice formation a sediment trap was deployed 0.5 m below the sea surface to capture the released brine droplets. The brine was collected in sample bottles in initially deflated plastic bags. The sampling interval was 40 min and seven samples were collected. During the time of active ice formation three samples with salinity above the background salinity were collected. To observe the buildup and ejection of the saline top layer a 25 m long drifting tube with temperature and conductivity sensors at 2.5 and 22.5 m was deployed. The experiment only lasted about six hours, but a gradual increase and a sudden lowering of the salinity was observed at the upper instrument but not seen at 22.5 m. This could perhaps indicate a buildup and release of a salinity anomaly (Quadfasel and Rudels, 1990; Rudels and Quadfasel, 1991).

With the advent of the CTD the resolution and accuracy of the salinity and temperature measurements increased tremendously and unexpected details were observed in the Greenland Sea water column. θS structures on stations taken in the Greenland Sea in 1993 by FS Valdivia are shown in Fig. 12. The densest water is the Greenland Sea bottom water concentrated in a narrow θS range with temperature and salinity decreasing towards the bottom. Above the bottom water the salinity range increases with increasing temperature and two salinity maxima can be identified. The highest salinities are found over the Greenland continental slope, along the Jan Mayen Ridge and at the Mohn–Knipovich Ridge and the lowest at the center of the Greenland Sea gyre. This implies that saline deep waters are advected from north along the Greenland slope and become deflected eastward to circulate around the rim of the Greenland Sea.

The saline deep waters must ultimately arrive from the Arctic Ocean and Fig. 12 also shows θS curves from stations in the Arctic Ocean obtained by the icebreaker Oden in 1991. Two saline deep water modes are clearly seen. Warmer and saline deep and bottom water are observed in the Makarov Basin close to the Lomonosov Ridge. Some of this deep water crosses the sill (1800 m) of the Lomonosov Ridge and enters the Amundsen Basin, where it can be identified as a salinity maximum around 1700 m (Anderson et al., 1994; Björk et al., 2007). The deep and bottom waters formed in the Eurasian Basin are colder and denser but slightly less saline. The density of the Eurasian Basin Deep water corresponds well with the deeper salinity maximum observed in the Greenland Sea, while the density of the upper salinity maximum, also forming a deep temperature maximum in the central Greenland Sea, indicates the presence of the Amerasian Basin deep water.

These observations show that the deep waters in the Arctic Mediterranean Sea are supplied by two sources. The open ocean convection area in the Greenland Sea provides the colder, less saline endmember and the Arctic Ocean the saline, warmer endmember. The deep waters communicate through Fram Strait and depending on the regional climate, one or the other source will dominate (Aagaard et al., 1985; Rudels, 1986a).

The salinity maxima are weakest in the central part of the Greenland Sea as is seen in Fig. 13 taken by RV Lance 1988, where an eddy exchange between the center and the rim of the Greenland Sea is captured. The exchange is isopycnal and the profiles show the doming of the density field in the central Greenland Sea, shifting the water from the rim upwards and that from the center down to deeper levels at the rim. It is thus likely that the saline Arctic Ocean deep waters that penetrate to the inner part of the Greenland Sea become diluted by the convection taking place there, largely removing their temperature and salinity anomalies.

Figure 14 displays the potential temperature and salinity profiles and the θS curves of two stations taken in the center of the Greenland Sea in 1988, station 65 in February and station 33 in June and a third station, 13, taken in 1993, all from FS Valdivia. The February station shows convection deeper than 1000 m and the formation of a homogenous, less saline layer covered by cold, low salinity surface water and located above a more saline deep layer. In June the profiles suggest that convection has continued and reached deeper than 2000 m. The salinity in the upper 1000 m has increased and the salinity in the deeper part has decreased, implying that during late winter the upper layer density has increased and the two layers have been homogenized. The density of the low salinity surface layer has to increase, first by cooling to freezing temperature and if this is not sufficient for convection, ice formation and brine rejection continue to increase the density to above that of the layer below, and the surface layer then convects. The density of the first layer below has then increased sufficiently for it to convect into the second deep layer below. The two layers then merge, forming a more than 2000 m deep, homogenous water column.

Station 13 in Fig. 14 shows a warmer, more saline water column and deep coinciding temperature and salinity maxima. The convection here extends to the temperature maximum and the convected water forms a temperature minimum above the temperature maximum. The deeper salinity maximum has become more saline and the Greenland Sea bottom water salinity has increased. That the convection is shallower implies that the Arctic Ocean deep water will not be transformed in the central Greenland Sea and the deep salinity maxima become more prominent. This especially applies to the deeper maximum from the Eurasian Basin.

Figure 15 shows two stations, one from 1982 taken by RV Hudson and one Valdivia station from 1993. The temperature and salinity of the deep salinity maximum have increased and also the Greenland Sea bottom water has become warmer and more saline. The salinity increase in the bottom water can only be caused by turbulent mixing from above and Meincke et al. (1997) estimated the vertical diffusion coefficient for the deep mixing in the Greenland Sea by assuming a balance between the lateral advection of Eurasian Basin deep water to the center of the gyre and the vertical mixing into the bottom water. The obtained mixing coefficient was surprisingly high, κv≈ 3.7 × 10⁻³ m2s-1. The renewal time for the deep salinity maximum was estimated to 6 years, corresponding to a through flow of 0.3 Sv. If the removal of the Greenland Sea bottom water, 0.2 Sv, is included 0.5 Sv of deep water is exported from the Greenland Sea (Meincke et al., 1997). Later studies made by Somavilla et al. (2013) and Marnela et al. (2016) have indicated similar high vertical mixing coefficients.

The changes in the Greenland Sea water column continued during the 1990s and a temperature maximum started to develop at mid depth (Budéus et al., 1998), indicating that the Amerasian Basin deep water penetrating towards the center of the Greenland Sea was no longer removed or weakened by mixing with convecting cold, less saline Arctic intermediate water. The water locally formed in winter has become less dense and only the water column above the temperature maximum is ventilated by convection. The temperature maximum, by contrast, is renewed by advection, locally from the Greenland slope but ultimately from the Arctic Ocean, which gradually increases its temperature. The temperature maximum also moved downwards in the water column (Budéus and Ronski, 2009), implying that the local convection, although less dense than the water of the temperature maximum, did not stop sinking when it reached its neutral density level. Instead, its layer thickness increased, indicating that it was pressing the underlying water column downwards, by continuity forcing the bottom water towards the rim and eventually to leave the Greenland Sea.

This radically changed the density structure of the central Greenland Sea. Instead of displaying a dome of dense water reaching towards the surface the density surfaces formed a bowl depressed in the center and rising towards the rim. This change implies that the doming, which was earlier considered generated by wind forcing and a necessary preconditioning for convection (Killworth, 1983; Marshall and Schott, 1999) instead is a consequence of open ocean convection. When deep convection ceases, the doming cannot be maintained and the density surfaces relax.

During this phase deep reaching cold, vertically almost homogenous anticyclonic vortices were observed in the Greenland Sea (Gascard et al., 2002; Wadhams et al., 2002). Their lateral extent was 10 to 15 km, close to the local Rossby radius, which prevented them from breaking up. The vortices reached deep (> 2000 m) and penetrated into the temperature maximum. The upper part of the vortices was denser but their lower part less dense than the ambient water column. The vortices were, as the rest of the Greenland Sea, capped by a low salinity layer in summer. However, the large vertical extent of the vortices made this upper layer less thick and it would more rapidly become dense in winter and convect into the vortices, recharging them with dense water. This early started and confined convection could also be a cause for the deep vertical penetration of the vortices. The fact that the vortices were less dense than the ambient water between 1500 and 2000 m made them, however, unlikely candidates for a renewal of the Greenland Sea deep and bottom water, but they stand out as intriguing features in the water column. Vortices were observed during a 10 years period but then disappeared.

In 1996 a tracer release experiment to study the convection was conducted in the central Greenland Sea. A cloud of Sulphur Hexachlorid was injected in the central gyre at the density surface σ0.5 = 30.4268 and its spreading and mixing were monitored during the following years (Watson et al., 1999). The first winter was one with comparably strong ice formation, which likely influenced the mixing of the tracer. Outside the gyre the tracer cloud spread but little vertically but in the gyre the tracer was redistributed vertically from the surface down to 1300 m. The highest tracer concentration was reduced and the maximum was shifted slightly upward.

One interpretation of these changes is that cooling or ice formation and brine rejection create denser water at the surface that convects into the deep as entraining plumes that pass through the tracer cloud and redistribute the tracer to deeper levels. By mass continuity a compensating upward flow through the tracer cloud then brings the tracer also to shallower levels. The one-dimensional convection model applied by Watson et al. (1999) agrees with this picture.

With time the temperature of the cold Arctic intermediate water above the temperature maximum gradually increased and eventually it became higher than the temperature of Amerasian Basin water, which instead only was identified as a sharp salinity increase below the ventilated intermediate water. By mid 2010s the ventilated intermediate water had also become more saline than the Amerasian Basin water and its salinity maximum disappeared.

This is a likely consequence of the increase in salinity and temperature of the upper layer in the Greenland Sea due to interaction with recirculating Atlantic water in the East Greenland Current, which counteracts the atmospheric cooling taking place in the central gyre (Latarius and Quadfasel, 2016). The salinity and temperature of the upper layer in the Greenland Sea are now the highest observed in the Greenland Sea water column. This has made the Greenland Sea similar to the Labrador Sea and the convection now resembles that of a gradual cooling and deepening of the upper layer in winter.

The cold, less saline Greenland Sea bottom water has also disappeared and the θS characteristics of the bottom water are now those of the Eurasian Basin deep water (Marnela et al., 2016). The features of the water column that identified two competing sources of the deep water in the Arctic Mediterranean (Aagaard et al., 1985; Rudels, 1986a) have disappeared, and presently the Arctic Ocean source dominates and the Greenland Sea deep water is renewed by advection. If the Arctic Ocean source also has weakened, it has not yet been detected in the Greenland Sea because the larger reservoir present in the Arctic Ocean. The cause behind these changes is likely the warming climate that no longer creates water dense enough to ventilate the deep Greenland Sea by open ocean convection.

The Arctic Ocean and Greenland Sea deep waters are mostly confined to the Greenland Sea by the Jan Mayen fracture zone and the Mohn–Knipovich Ridge and contribute but little to the overflows. When the intensive studies of the Nordic Seas started in the 1970s and 1980s an outflow of Greenland Sea deep water to the Norwegian Sea was observed in the about 2200 m deep channel in the Mohn Ridge just north of Jan Mayen. In the late 1990s this flow had changed direction and instead deep water from the Norwegian Sea entered the Greenland Sea, implying that the Greenland Sea deep water density had been reduced and that the deep pressure gradient had reversed (Østerhus and Gammelsrød, 1999). Since no deep water is formed in the Norwegian Sea, this indicates that water, earlier exported from the Greenland Sea, was returning. This situation probably persisted only a short period since, due to continuity, the deep water from the Arctic Ocean entering the Greenland Sea has to leave, either through gaps in the Mohn–Knipovich Ridge to the Norwegian Sea or return to the Arctic Ocean through Fram Strait.

The deepest passage across the Greenland–Scotland Ridge to the North Atlantic is the Faroe Bank Channel (> 840 m) and the densest deep water that enters the Norwegian Sea from the Greenland Sea must eventually exit the Arctic Mediterranean through this passage. For less dense water there is also the possibility of spillover across the 500 m deep ridge segment between Iceland and the Faroes. The Faroe Bank Channel is a continuation of the Faroe Shetland Channel and Fig. 16, showing bottom temperature and bottom salinities in the northeastern North Atlantic, reveals the narrowness of the channel and also the spillover of cold, less saline water across the Iceland-Scotland Ridge. Not all dense water that passes through the Faroe–Shetland Channel reaches the Faroe Bank Channel. Some less dense water continues south and crosses the Wyville Thomson Ridge and enters the Rockall Basin. As the name of the ridge suggests, this is where Carpenter and Wyville Thomson first observed the cold water entering the Northeast Atlantic from the north.

Not all the overflow water from the Norwegian Sea derives from the Greenland Sea. A substantial fraction of the East Greenland Current that crosses the Jan Mayen Fracture Zone and enters the Iceland Sea continues eastward in the East Icelandic Current into the Norwegian Sea. The East Iceland Current is deep, about 600 m, and carries not only low salinity surface water but also denser waters that especially contribute to the spillover across the Iceland-Faroe Ridge. In a drifter experiment with RAFOS floats in the Iceland Sea with a drifting depth of 600 m all of the floats deployed east of the Kolbeinsey Ridge moved southeastward into the Norwegian Sea, and also some deployed west of Kolbeinsey Ridge entered the Norwegian Sea. However, most of the floats deployed west of the ridge either exited through Denmark Strait or grounded on the northwestern Icelandic shelf (De Jong et al., 2017).

The eastern overflow entrains strongly the ambient waters and as the overflow reaches its neutral density level the overflow volume has more than doubled. The overflow water moves toward the equator but when it reaches the Gibbs Fracture Zone in the Mid-Atlantic Ridge it splits and some, maybe the major part, leaves the Iceland Basin and enters the Irminger Basin and then moves north towards Iceland. The initial low temperature and salinity characteristics of the eastern overflow has rapidly been removed and it has attained a potential temperature between 2 and 3 °C and a salinity around 34.925.

The western overflow in Denmark Strait is more complicated. It is not a deep, almost stagnant reservoir that gradually drains south over a sill, but instead a region where several current branches and water masses converge. The upper part of the East Greenland Current that passes over the sill depth of the Jan Mayen Fracture Zone (1700 m) comprises not only low salinity Polar water, but also Arctic Atlantic and denser intermediate waters from the Arctic Ocean, as well as warm and saline Atlantic water recirculating in Fram Strait and Arctic Intermediate waters created in and exported from the Greenland Sea. There is also the recently discovered North Icelandic Jet roughly following the 600 m isobath on the northern Iceland slope westward to Denmark Strait (Jónsson and Valdimarsson, 2004). In addition, the warm and saline North Icelandic Irminger Current enters the Denmark Strait from the south. The sill depth in Denmark Strait is 640 m, but the deep channel is narrow and waters from north and south contend for space and passage. Changes take place from one week to the next and Fig. 17 shows two sections along the sill taken by RV Aranda one week apart. On the first section waters from the north dominates, on the second water from the south. The second section extended farther towards Greenland because of favorable ice conditions (Rudels et al., 1999c).

Overflow water is commonly defined as water with potential density above 27.8 and the second section shows that waters with density above 27.8 are present on the shelf, which implies that even when the North Iceland Irminger Current dominates in the deep channel the overflow, or at least part of it, could instead take place over the shelf and then sink down the continental slope into the Irminger Basin. However, the East Greenland Current is not the only source of overflow water. In fact, the densest water crossing the sill is supplied by the North Icelandic Jet. The jet has been traced eastward towards its unknown source area and it is observed along the entire northern slope (Våge et al., 2011).

Våge et al. (2011) suggested that the water of the North Icelandic Jet is supplied by two sources, the first would be the cooling of the North Icelandic Irminger Current as it flows eastward on the shelf. The second contribution would come from mixing with the denser water in the interior of the Iceland Basin as the cooled water returns towards Denmark Strait, further increasing its density. The problem with this explanation is that the atmospheric cooling observed in the area is not strong enough to create the observed θS properties of the jet (Segtnan et al., 2011; Latarius and Quadfasel, 2016). Furthermore, the seasonal convection in the Iceland Sea is usually limited to the upper 200 m and only in the northern part of the Iceland Sea does the winter mixed layer occasionally attain the θS observed in the North Icelandic Jet (Våge et al., 2015).

That the formation of dense water in the Iceland Sea in winter and its subsequent export in summer and fall was the main process supplying the Denmark Strait overflow was proposed by Swift et al. (1980) and Swift and Aagaard (1981) as an alternative to the cooling of the Atlantic water in the Norwegian Sea and its recirculation in Fram Strait and export in the East Greenland Current suggested by Worthington (1970). However, Mauritzen (1996a, b) did not find any strong seasonality in the overflow, which speaks against seasonal sources in the central basins of the Nordic Seas and rather indicates a steady flow, which could be supplied by the gradually cooled Atlantic water returning to the North Atlantic through Denmark Strait.

Waters with the θS characteristics of the deepest overflow are present in the northern Iceland Sea and in the East Greenland Current. These θS properties are, however, observed at much deeper levels, 750 to 1500 m, than the about 500 m depth, where the core of the North Icelandic Jet is found. Moreover, it is likely that these waters are formed farther north and have been advected across the Jan Mayen Fracture Zone to the Iceland Sea. If the East Iceland Current transports a major part of the upper 600 m of the water column to the Norwegian Sea, the deeper levels might, if they are advected southward, ascend to shallower depth as they approach the Iceland Continental slope and then feed the North Icelandic Jet (Fig. 18).

As a curiosity, the deep salinity maxima shown in Fig. 18 derive from the Jan Mayen Fracture Zone (blue station) and the central Greenland Sea (red station) and have about the same density, but the maximum at the station at the rim is located at shallower depth than the maximum at the station in the central Greenland Sea. This implies that the convection in the Greenland Sea has compressed the underlying water column, creating the deep part of a bowl, which then rises towards the rim. These stations were taken in 2002, when the Greenland Sea convection had become shallower, and it should be compared with the stations in Fig. 13 in Sect. 10, taken in 1988, when the denser water is found at shallower levels in the center than at the rim.

The overflow plume in Denmark Strait is stratified. It comprises a cold, < -0.5 °C, bottom layer with salinity ≈ 34.9 below a temperature and a salinity maximum of 2 °C and 34.91, respectively, and it is capped by a salinity minimum, a low salinity lid (Fig. 19). As the plume leaves the sill its bottom temperature increases rapidly to above 0 °C but remains below 1 °C even when the plume has descended deeper than 3000 m in the Irminger Sea, and the low salinity lid is present almost as deep. The initial rapid increase in bottom layer temperature suggests that a direct mixing between the overflow water and the northward flowing North Iceland Irminger Current could take place already at the sill in Denmark Strait. There the deepest and densest part of the overflow comes in almost direct contact with the warm and saline Irminger Current water, and the observed increase in plume temperature in the deep Irminger Basin could result from a mixing ratio at the sill of one part of entrained water to four parts of overflow water. This is significantly less than if the entrainment would occur deeper down the slope. It is also less than the estimated entrainment for the Faroe Bank overflow.

The temperature minimum and the low salinity lid, by contrast, must derive from the Greenland shelf, where waters of the observed properties are found. The fact that especially the low salinity lid survives as the plume descends deeper than 2000 m also indicates that the entrainment of ambient water through the upper boundary of the plume is weak. However, the low salinity lid suggests that if the less dense water can remain in contact with the densest part of the plume in the early stages, when it starts to descend the slope, it will remain attached to the plume and sink into the deep. One factor that can facilitate the descent of the cold low salinity upper part of the plume is the thermobaric effect, which increases the relative density of colder water as the pressure increases. As the Denmark Strait plume descends it eventually sinks through the overflow water from the Faroe Bank channel that shows up as a salinity maximum above the colder, less saline Denmark Strait overflow. Both overflows then continue to the Labrador Sea.

If the dense water that flows southward on the Greenland shelf reaches the shelf break farther south, where the shelf narrows, the entrainment appears to be stronger and the plumes will remain on the upper part of the slope and not connect to the deeper overflow. This could be the case for the Greenland spill jet, identified by the density surfaces rising towards the shelf, which indicates velocities increasing with depth (Pickart et al., 2005). It flows as a separate core above the deep overflow plume towards Cape Farewell and eventually the density surfaces flatten and the spill jet disappears (Pickart et al., 2005).

The overflow in Denmark Strait is strong and varies on time scales of a few days (2 to 5 d) The strong overflow events were noticed by Worthington (1969) as the first direct current meter array was recovered. These strong overflow events had periods slightly less than a week. However, later observations (Von Appen et al., 2017) have shown that two different kinds of strong overflow events are present. The boluses are related to overflow events where the thickness of the overflow plume increases drastically, the temperature of the boluses decreases, but no significant increase in the overflow velocity is observed. The second overflow anomaly Von Appen et al. (2017) denoted as pulses. The pulses show an increase the overflow velocity but the thickness of the overflow layer decreases. The temperature reduction is less (-0.1 °C) than that observed for the boluses (-0.4 °C) and the more rapid flow in the pulses compensates for the thinner overflow plume and the overflow strength is about the same for the boluses and the pulses.

The difference of height of the overflow plumes first suggests a supercritical flow for the boluses and a subcritical flow for the pulses, but nature is most likely more devious than this. The flow is rotating and the thickening of the boluses is associated with cyclonic rotation and the thinning of the pulses relates to anticyclonic rotation (Von Appen et al., 2017). The most interesting question is, however, if the boluses and pulses are associated with different water masses. The almost similar temperature of the overflow water in the two cases suggests that the main sources are in the central gyres, not in the East Greenland Current, and rather in the Greenland Sea than in the Iceland Sea.

The overflows contribute together with the dense water formed by open ocean convection in the Labrador Sea to the North Atlantic deep water and to the Atlantic Overturning Circulation. The overturning circulation has been monitored since 2004 by the RAPID-MOCHA (Meridional Overturning Circulation and Heat-flux Array) array deployed at 26° N. One of the drawbacks with this array is that it is located too far south to separate the overturning circulation, sinking into the deep, from the horizontal recirculation of cooled upper water in the subtropical gyre. An additional array, OSNAP (Overturning in the Subpolar North Atlantic Program), located north of the subtropical gyre has therefore been proposed and accepted, and the first deployment was in 2014. The array extends from Scotland to Greenland (OSNAP east) and from Greenland to Labrador (OSNAP west). The first deployment ended in 2016 and when the results from the first 20 months were published (Lozier et al., 2019) the title “A sea change in our view of overturning in the subpolar North Atlantic” suggests that the results, at least for some of the researchers involved, have come as a surprise. The convection and dense water formation in the Labrador Sea contribute, as estimated by the OSNAP west array, but little, 2.1 ± 0.3 Sv, to the overturning circulation. By contrast the transformation to denser water measured by the OSNAP east array was 15.6 ± 0.8 Sv, almost an order of magnitude larger. Since some recirculation takes place, the measured total overturning is somewhat less, 14.9 ± 0.9 Sv.

For those who have been working in the Arctic Mediterranean Sea and on the exchanges across the Greenland–Scotland Ridge these numbers are not surprising and should also be familiar to those remembering L.V. Worthington's (1976) monograph on the North Atlantic circulation. The observations made during the NACLIM project (North Atlantic CLIMate) indicated an overflow of 5.8 Sv (Østerhus et al., 2019). Assuming a doubling of the volume due to entrainment into the plumes and some contribution from the spill jet the numbers almost match.

The small volume of Atlantic deep water formed in the Labrador Sea comes as a surprise though. It suggests that seasonal deep convection is not a process that creates and drives an overturning circulation. It forms MODE waters rather than deep water (McCartney, 1977; McCartney and Talley, 1982). To create the dense water that supplies the overturning circulation, a sill and an isolated basin might be required. This would inject water with substantially higher density contrast that can sink down and entrain ambient water, creating a significant volume of deep water with distinct properties. The fact that also the much weaker Mediterranean overflow shows up in and characterizes the Atlantic water column over a large region suggests that this could be the case.

Observations from the RAPID array indicates a reduction of the overturning circulation by 2.5 Sv in 2009. The weaker overturning has persisted but no further reduction has been observed (Bryden et al., 2020). The reduction has, surprisingly, taken place in the denser part of the overturning circulation, which is supplied by the overflow from the Arctic Mediterranean. During the same time the EU projects THOR (ThermoHaline Overturning at Risk) and NACLIM were running and no reduction was observed, not in the overflows, nor in the northward inflows of Atlantic water to the Arctic Mediterranean.

The RAPID array showed that reduction of the overturning circulation was compensated by an increase in the recirculation in the subtropical gyre by 2.5 Sv, about 15 % of the overturning circulation. This implies that less warm Atlantic water is transported by the North Atlantic drift to the subpolar gyre, which should lead to a cooling of the northeast Atlantic.

The water leaving the Arctic Mediterranean Sea consists not only of dense overflow water but also of cold, less saline and less dense polar water that interacts with the upper part of the subpolar North Atlantic. There are two main, interacting, outflows, the outflow of Polar Water through Fram Strait in the East Greenland Current that follows the Greenland shelf and coast southward through Denmark Strait to Cape Farewell, and the polar water that flows through the narrow and shallow passages in the Canadian Arctic Archipelago to Baffin Bay and then continues through Davis Strait to the Labrador Sea. These two flows interact. A part of the East Greenland Current turns north after it has rounded Cape Farewell and forms the West Greenland Current that enters Baffin Bay from the south. The upper, polar, part of West Greenland Current mixes with the outflows from Nares Strait, Jones Sound and Lancaster Sound and the mixing product is transformed by ice formation during winter into a distinct, about 200 m deep upper layer with temperature of -1.6 °C and salinity of 33.7 (Fig. 20).

Below the low salinity upper water, a warm, 3–4 °C, but rather fresh, 34.5–34.6, Atlantic layer is present. The Atlantic water enters Baffin Bay through Davis Strait in the West Greenland Current and the low salinity indicates that it was originally a part of the Irminger Current that has become diluted by cold, low salinity water from the East Greenland Current south of Denmark Strait.

The bottom water in Baffin Bay is again colder, about -0.45 °C with salinity 34.45. The origin of the Baffin Bay bottom water has been debated, Sverdrup et al. (1942) considered it to be a mixture between Labrador Sea water and Baffin Bay surface water, and local formation by convection was suggested by Bourke and Paquette (1991). Bailey (1956) found TS characteristics at sill depth north of Nares Strait similar to those of the Baffin Bay bottom water and proposed that bottom water entered from the Arctic Ocean. Figure 21 shows the TS curves from one station in Baffin Bay and two stations from the Arctic Ocean, one from the continental slope in the Canada Basin and the other from the slope northeast of Greenland, indicating that the Baffin Bay bottom water could derive from the Arctic Ocean.

The TS curves also show the warm but less saline characteristics of the Baffin Bay Atlantic water compared to the Arctic Atlantic water to the north. They also, interestingly, indicate that the entrainment of ambient water into the deep inflow must be small. If the inflow entrains much, it will become warmer, which reduces its density. This, in turn, requires a denser initial inflow. However, that would also mean a warmer inflow, which would lead to different TS characteristics in the Baffin Bay bottom water.

To determine the export of low salinity polar water by direct current measurements is difficult. Drifting ice and bad weather conditions complicate not only the deployment and recovery but also the survival of the moorings. Since one primary object is to determine the freshwater fluxes that may influence the convection in the subpolar gyre, it is essential to obtain measurements close to the surface and the ice, where most of freshwater is located. In spite of these difficulties measurements of volume and freshwater fluxes have been made, at least for shorter periods, in all the major passages; Lancaster Sound (Peterson et al., 2012), Jones Sound (Cardigan Strait and Hells Gate) (Melling et al., 2008), and Nares Strait (Münchow and Melling, 2008; Rabe et al., 2010) in the Canadian Arctic Archipelago, in Davis Strait (Curry et al., 2011, 2014), and on the shelf area east of Greenland (De Steur et al., 2009; De Steur et al., 2016). The techniques used in the different passages are distinct and adapted to the local challenges such as the presence of sea ice and the closeness to the magnetic North Pole that complicates the direction measurements. Details of these observations are given in Melling et al. (2008) and in Rudels (2021).

The difficulty to perform field observation, especially for monitoring purpose, makes it tempting to use existing simple models to estimate the transports of Polar water through the different passages. One approach follows Werenskiold (1935), who derived an expression for the geostrophic transport M in a coastal current based only on the depth of the upper layer at the coast, H1, and the density difference (ρ2-ρ1) between the upper layer and the lower layer at rest, which extends to the surface one Rossby Radius away from the coast. 4M=(ρ2-ρ1)gH122fρ2

The density in cold Arctic waters is mainly controlled by salinity and can be written as ρ = ρr(1+βS), where ρr is the reference density and β the coefficient of salt contraction. Taking the density of the lower layer as the reference density the reduced gravity ρ′ becomes gβ(S2-S1).

This represents the upper circulation and the amount of lower layer Atlantic water that enters the Arctic Ocean and is entrained into the upper layer. Stigebrandt (1981) modelled the entrainment with a Kato–Phillips mixing model and noticed that if the openings are wider than a Rossby radius they all transport with the same rate.

There are three straits that are wider than the Rossby Radius and can transport at full capacity, Fram Strait, Lancaster Sound and Nares Strait. The total transport through these passages comprises the liquid freshwater flux F added to the water column by rivers and by the low salinity Pacific inflow, and the entrained lower layer water M2. The Pacific inflow is here separated into freshwater and lower layer water fluxes. This gives: 5M2+F=3gβ(S2-S1)H122f and the thickness of the freshwater m in the upper layer becomes: 6m=(S2-S1)S2H1

The salinity of the upper layer can be written as: 7S1=M2S2(M2+F)

And introducing this in the expression for the outflow gives: 8(M2+F)=3gβFS22f12H1 and m becomes: 9m=S2(M2+F)-(M2S2)S2(M2+F)H1=FH1(M2+F)=2fF3gβS212

The Rossby Radius Ro is independent of the entrainment and only depends on the freshwater input: 10Ro=[gβ(S2-S1)H1]12f=2gβS2F3f314

The Rossby radius is about 10 km, which confirms that both Lancaster Sound and Nares Strait are wide enough to transport with full capacity.

If we assume that the depth H1 is the same in all channels and equal to the depth of the upper layer in the Arctic Ocean, the liquid freshwater storage in the Arctic Ocean can be estimated. Introducing g = 9.83 ms-2, f = 1.4 × 10⁻⁴ s-1, β = 8 × 10⁻⁴, F = 0.20 × 10⁶ m3s-1 and setting S2 equal to 35 and the salinity of sea ice to zero the liquid freshwater storage becomes 8.24 m. The area of the Arctic Ocean is about 10 × 10¹² m2 and with a freshwater export of 0.2 Sv the residence time becomes 13 years, or half of this if only the deep part of the Arctic Ocean is considered. The freshwater storage increases as F1/2 but its residence time decreases as F-1/2 (Rudels, 2010).

The thickness of the ice cover was in the 1960s estimated to 3 m, about one third of the liquid freshwater storage but had decreased by about 40 % by the 1990s (Rothrock et al., 1999). If the ice export Fi = 0.09 Sv, remains constant, the mean residence time of the ice cover is between 2.6 and 5.3 years depending upon the area considered. If only a seasonal ice cover were present and no sea ice were exported the liquid freshwater storage would increase to a little above 10 m (Rudels, 2010). The liquid freshwater in the Arctic Ocean is, however, not a lid lying on top of a deeper water mass but is distributed over a large, spatially varying, depth range in the Arctic Ocean water column. The conditions in the three passages are also not equal. Especially the characteristics of the second layer differ between the passages, which affects the transports of both volume and freshwater.

The second passageway from the Arctic Ocean to the North Atlantic, the Canadian Arctic Archipelago and Baffin Bay, offers additional challenges but also other opportunities to theoretically estimate the transports. The geostrophic transports depend upon the number of openings (Stigebrandt, 1981) and two passages, Lancaster Sound and Nares Strait, transport Polar water from the Arctic Ocean to Baffin Bay. In addition, the West Greenland Current carries water, ultimately from Fram Strait and the Arctic Ocean, into Baffin Bay from the south. These inflows have to exit through one opening, Davis Strait, to the Labrador Sea.

The outflow from the Arctic Ocean is ultimately due to the higher sea level in the Arctic Ocean, which implies that the upper layer in Baffin Bay must be denser than that in the Arctic Ocean and its depth should be greater than the depth of inflows to Baffin Bay. There will be a pressure reversal in depth in Baffin Bay, below which the pressure gradient forces the Baffin Bay water to enter the inflow channels. The water from Baffin Bay does not enter the Arctic Ocean but it forms a lower boundary below the inflows that is taken as the depth of the transports to Baffin Bay. The depth and density of the Baffin Bay upper layer determine the geostrophic transport out of Baffin Bay, which should balance the northern inflows. The Arctic Ocean is connected to the deeper layers in Baffin Bay through Fram Strait and across the sills of the Greenland–Scotland Ridge and Davis Strait, removing any deep pressure gradient (Fig. 22).

Assuming that the temperature and salinity of the Baffin Bay upper layer are -1.6 °C and 33.7, the upper layer depth in Baffin Bay can be found, and the geostrophic transports through the system can be determined as functions of the Polar water salinity or density (Rudels, 1986b). The transports decrease with increasing salinity of the Polar water, since the density difference between the layers is then reduced.

However, the characteristics of the Baffin Bay upper layer are not only due to the mixing between the entering waters but also influenced by the cooling and ice formation taking place in winter. The cooling and the ice formation only depend on the atmospheric forcing and are independent of the inflows, and a higher salinity of the entering Polar water would require an increased transport into Baffin Bay to reproduce the observed salinity of the Baffin Bay upper water.

Since one estimate gives a transport decreasing with increasing salinity and the other an increasing transport with increasing salinity, a salinity must exist for which these transports are equal. This salinity is then set as the salinity of the Polar water leaving the Arctic Ocean. The Polar outflow, the freshwater export, and the ice export through the Canadian Arctic Archipelago–Baffin Bay system can then be determined (Rudels, 1986b). However, the outflow estimate obtained by this approach, 0.7 Sv, was lower than the 2 Sv then commonly cited. One serious weakness of this approach is that it requires a residence time for the Polar water in Baffin Bay of at least one year for the transformation into the Baffin Bay upper layer to take place. Especially the inflow through Lancaster Sound that enters close to Davis Strait may pass through Baffin Bay into the Labrador Sea in a time shorter than 1 year.

The expected change into a warmer climate has brought melting of the Greenland ice cap into focus. Intensified melting has been observed in the Jacobhavn isbrae in Baffin Bay, but also the low salinity East Greenland Coastal Current (Sutherland and Pickart, 2008), flowing on the Greenland shelf inshore of the East Greenland Current, indicates that freshwater is supplied from Greenland as runoff and as ice melt. This freshwater input, either directly to Baffin Bay and/or advected through Davis Strait into Baffin Bay, will lower the salinity of the upper layer in Baffin Bay.

The density contrast between the Polar water and the Baffin Bay upper layer then diminishes and the transport through the Archipelago becomes smaller. The expected increase in mass and freshwater fluxes from the Arctic Ocean to the North Atlantic in a warmer climate, might then not take place through the Canadian Arctic Archipelago. Instead, much of this increased outflow could be channeled through Fram Strait, which might affect the dense water formation and convection both in the Greenland Sea and in the Irminger and Labrador seas (Rudels, 2011).

It should be noted that in Rudels (2011) the ice formation and ice export are computed from the mass and salt balances in Baffin Bay using the postulated salinity of the Baffin Bay upper layer. It is then not independent of the inflow salinities. This in contrast to the approach in Rudels (1986b), where the ice formation was considered due to the atmospheric heat loss and unrelated to the salinities of the transports. It could then be used, together with the salinity of the upper layer, to determine the salinity of the Polar water entering Baffin Bay. The author appears, 25 years later, to have forgotten this aspect.

The Baffin Bay water column comprises not only transformed Polar water but also Atlantic water advected from the south and cold deep and bottom waters originating in the Arctic Ocean. These waters must eventually leave Baffin Bay, which requires a deep circulation in Baffin Bay that has not, so far, been considered. Rudels (2011) tried to theoretically estimate this deep circulation and how it relates to the upper layer exchanges by assuming that the inflows of upper water through the northern channels and in the West Greenland Current, here considered as part of the Polar inflow, are shallower than the depth of the pressure reversal.

This allows for inflows from the deeper layers to Baffin Bay. Furthermore, the pressure reversal between the Arctic Ocean and Baffin Bay does not occur in Baffin Bay but in the Labrador Sea, below the sill depth in Davis Strait. The pressure head due to the higher sea level in the Arctic Ocean is thus not removed in Baffin Bay and it forces not only a deep inflow in the north but also a deep outflow to the Labrador Sea. The pressure reversal in the West Greenland Current, by contrast, occurs above the sill depth and the deep Atlantic inflow recirculates in Baffin Bay and then exits through Davis strait together with, but above, the outflow of the deep waters from the Arctic Ocean (Rudels, 2011).

The deep inflows decrease with increasing salinity of the Polar water, but the deep outflow through Davis Strait increases with the salinity of the Polar water. This implies that there is a salinity, 33.2 in this case, of the Polar water for which the deep exchanges in Baffin Bay are balanced. The deep outflow from the Arctic Ocean, taking place in the deepest channel, Nares Strait is 0.62 Sv, which is almost as large as the Nares Strait upper layer outflow of 0.77 Sv (Münchow and Melling, 2008). The added barotropic pressure gradient also increases the outflow of Polar water from the Arctic Ocean to the North Atlantic and the total outflow through the Canadian Arctic Archipelago–Baffin Bay system becomes about 2.2 Sv, somewhat larger than the about 1.5 Sv presently determined from observations (e.g., Curry et al., 2014). For details of this theoretical approach the reader is referred to Rudels (1986b, 2011).

All waters that circulate around the Arctic Mediterranean eventually flow south, entering the subpolar gyre in the North Atlantic. For most waters it is a return to where they started from as warm, saline Atlantic inflows crossing the Greenland–Scotland Ridge. Only the Pacific inflow and the freshwater input are added to the water entering the subpolar gyre from the north. We may remark here that in spite of the Pacific inflow bringing low salinity water to the Arctic Mediterranean, it constitutes a salt input to the Atlantic Ocean that eventually crosses the equator and is exported to the Southern Ocean.

The Atlantic water entering the Arctic Mediterranean becomes transformed in many ways by several different processes but, essentially, it is separated into denser and less dense waters by cooling and by receiving or rejecting freshwater, mainly through sea ice melting or sea ice formation. Waters that return to the North Atlantic either as dense overflows or buoyant outflows.

Østerhus et al. (2019) compiled the measured volume transports through all different control sections to find if the volume fluxes balance (Fig. 23 and Table 1). They found an inflow of 9.1 Sv and an outflow of 9.5 Sv, the exchange being almost in balance. However, an imbalance of 0.1 Sv would in 1 year lead to a sea level change in the Arctic Mediterranean of 0.2 m. This is more than that observed and indicates that the direct current measurements may still be improved.

However, what do not balance are the freshwater transports. Haine et al. (2015) estimated the freshwater storage in the Arctic Mediterranean in two periods, between 1980 and 2000 and between 2000 and 2010 (Table 2) and discussed the reason for the changes. What is clear, is that the adjustment time for the baroclinic pressure field is much longer than that for the barotropic pressure field. Perhaps it will never really be in balance.

Of the Atlantic water about 6 Sv are transformed to dense overflow water while slightly more than 2 Sv become less dense water that together with the Pacific water make up the outflow waters (Østerhus et al., 2019). This is slightly more than the 1 Sv of less dense Polar water and the overflow of 1.8 Sv plus 3.2 Sv of overflow water estimated by Brown et al. (2025) and Bacon et al. (2022), assuming that the Atlantic water transformed in the Nordic Seas is dense enough to become overflow water.

The overflow water sinks into the deep and eventually joins the southward moving deep western boundary current, but the outflow water circulates in the upper part of the subpolar gyre and will influence the upstream conditions of the inflows to the Arctic Mediterranean.

The most conspicuous northern input to the subpolar gyre was the “Great Salinity Anomaly” observed in the late 1960s and early 1970s (Dickson et al., 1988). The outflow of low salinity water and ice through Denmark Strait strongly reduced the salinity in the gyre and the anomaly could be traced around the subpolar gyre and then into the Nordic Seas at least as far as Fram Strait (Dickson et al., 1988).

Up to the mid-1990s the surface waters in the subpolar gyre were colder and less saline than their average temperature and the same conditions prevailed in the Nordic Seas and the Barents Sea. In the late 1980s to the mid-1990s the North Atlantic Oscillation (NAO) index, the normalized air pressure difference between the Azores and Iceland, was high and positive, and the cold continental air that dominated over the Labrador Sea led to a strong and deep convection and the created Labrador Sea water was exceptionally cold and fresh (Lazier et al., 2002). The cooling of the Labrador Sea accelerated the circulation of the subpolar gyre, which extended far to the east and partly prevented warmer water from the subtropical gyre to enter the Nordic Seas. The high NAO index was also manifested in the changes in the Barents Sea and in the Arctic Ocean, where the river runoff was diverted farther to the east, from the Amundsen Basin to the Makarov Basin.

As the NAO index eventually became lower the subpolar gyre relaxed and retreated, leaving a freer passage for the waters of the North Atlantic Drift to reach the Greenland–Scotland Ridge and enter the Nordic Sea (Hátún et al., 2005; Häkkinen et al., 2008). After this a period of almost 15 year of warm, saline surface water was observed in the subpolar gyre and in the Nordic Seas and the Barents Sea.

The North Atlantic drift splits as it approaches the European continent and one part turns south and remains in the subtropical gyre. The simpleminded picture would be that it is the southernmost branch that turns south, but observations from floats launched in the upper 15 m indicated that most of the floats turned south and instead the deeper layers, around 300 m and deeper, enter the subpolar gyre (Brambilla and Talley, 2006; Burkholder and Lozier, 2011). On its path towards the Nordic Seas the water from the subtropical gyre is also cooled and freshened by mixing with the low salinity upper layers of the subpolar gyre flowing eastward from the Labrador Sea (Burkholder and Lozier, 2011).

In 2009 the RAPID Array at 26° N showed an increase of 2.5 Sv in the southward recirculation of the North Atlantic Drift (Smeed et al., 2014). This would reduce the warm water inflow to the subpolar gyre and less warm water would enter the Norwegian Sea. This change probably ended or at least contributed to the end of the 20 years period of warm surface water in the subpolar gyre (González-Pola et al., 2023).

In 2012 a strong freshening of the subpolar gyre was observed, the strongest in 120 years (Holliday et al., 2020). This freshening appears to be different from the one associated with the Great Salinity Anomaly in that it originated in a rerouting of the circulation of the low salinity Labrador Current. Instead of moving south around Newfoundland, it was detached from the continental slope and joined the subpolar gyre circulation eastward. This low salinity signal has continued northward, into the Barents Sea and to Fram Strait (González-Pola et al., 2023). If this freshening, taking place at least in the surface water, will imply a stronger stability of the Arctic Ocean water column and perhaps limit the vertical mixing in the Arctic Ocean and possibly affect the ice–ocean interactions, remains to be seen.

The atmospheric circulation patterns and the wind fields thus strongly influence the circulation into and out of the Arctic Mediterranean Sea, but are they also the main causes for the overturning circulation? The overturning requires that the northward flowing water becomes denser and sinks into the deep. Moreover, it also appears to require a sill that allows the convection to build up a reservoir of denser water and create a density difference between the deep waters on the two sides of the ridge. As the dense water crosses the ridge, the density difference allows it to accelerate and entrain ambient water, practically doubling its volume. This is different from the deep convection in the Labrador Sea that increases its depth and its density only gradually and marginally relative to the underlying waters. This might be the main reason for the small contribution from the Labrador Sea to the overturning circulation.

The formation of dense water in the Arctic Mediterranean also provides a constant, but perhaps weaker and often less obvious forcing than the wind driven circulation. The dense outflowing water has to be replaced, when the sea level starts to become lower, if not before. The outstanding question is: Can the processes in the Arctic Mediterranean produce enough dense, and dense enough, water in the future? If the convection does not reach to the bottom, it will create stagnant conditions in the deep basins, but as long as intermediate waters are formed, still denser than the waters in the subpolar gyre, and the pressure head at the sill is present, the overturning circulation should be maintained. A large input of freshwater to the Greenland Sea, either from the Arctic Ocean or from ice melting on Greenland, could strongly affect the convection in the Greenland Sea (Stigebrandt, 1985). Only the cooling of the Atlantic water in the Norwegian Sea and the unknown rate of deep water formation in the Arctic Ocean would then remain as sources of dense overflow water, but that might be sufficient for maintaining the overturning circulation.

At the global scale the Arctic Ocean is a fundamental node in the global hydrological cycle and the ocean's thermohaline circulation (Fig. 24). Here we have assessed the system's circulation, water mass transformations, and key functions and processes. Our use of the term Arctic Mediterranean acknowledges that the Arctic Ocean and Nordic Seas are largely surrounded by north-flowing drainage basins supplied by moisture originally derived from subarctic seas and delivered by the Westerly wind field (cf. Carmack et al., 2008, 2016). As such they are termed beta oceans, where stratification is controlled by salinity which, in turn, allows ice formation and attendant seasonal fluxes of salt and freshwater (Carmack, 2007). The Arctic Mediterranean is, itself, dynamically and thermodynamically, a double estuary, in that its inflowing subarctic waters are transformed into outflow waters that are both lighter and denser than their parent water masses (Carmack and Wassmann, 2006; Rudels and Carmack, 2022; Brown et al., 2025), all flowing along the east and west coasts of Greenland. There they mingle and mix with new incoming subarctic waters to assemble the lower limb of the Atlantic Meridional Overturning Circulation.

The Arctic climate system, is now primed to cross multiple cascading and irreversible tipping points, as was forecast by Aagaard and Carmack (1989) who speculated, “that the present-day Greenland and Iceland seas, and probably also the Labrador Sea, are rather delicately poised with respect to their ability to sustain convection.”