Large-scale climate changes throughout the Cenozoic (from 65 Ma to present) have been linked to ocean gateway configurations and CO2 changes and their respective impact on the ocean circulation [1,2,3]. One of the major challenges of modeling past climate and ocean circulation is due to large uncertainties in boundary conditions during the geological past. Of the tectonic boundary conditions, the changes in ocean gateway geometries (i.e., Fram Strait; FS) have imposed significant changes in the Arctic Ocean and global ocean circulation [4,5,6,7].

The formation of Arctic–Atlantic gateways plays a key role in global climate history by driving heat transport between the Atlantic and Arctic Ocean [3]. The evolution of the FS, located between Svalbard and Greenland, is supposed to play a crucial role in ocean dynamics of the Arctic Ocean and, therefore, in the evolution of the north polar region [8]. The timing for the opening of FS is well constrained by magnetic and seismic data that show the FS was likely a narrow and shallow gateway around 21 Ma [9]. The early Miocene opening of FS [9] allowed an increased water mass exchange between the Atlantic Ocean and the Arctic Ocean, with significant climatic impacts that strongly influenced the paleoceanographic conditions in the Arctic Ocean [6], and likely caused enhanced contribution of the North Atlantic thermohaline circulation [11]. During this period, saline Atlantic waters entered to the Arctic and supplied oxygen to intermediate/deep waters transforming the Arctic Ocean from a restricted to a fully ventilated Ocean at around 17.5 Ma [8]. This also enhanced the heat transport and moisture supply to northern higher latitudes via the westerlies [12] and enabled extensive growth of floating ice shelves or sea-ice beginning at ~ 15 Ma [5].

As mentioned above, the geometry of the gateway (width and depth) and a deep-water circulation around 1500 mbsl across the FS [13] are critical parameter controlling the transformation from the closed Arctic Ocean to a ventilated basin [9]. Today, the FS width is ~ 670 km [3, 7] and it depth varies from 2550 to 2800 m [6, 7]. Using a two-layer analytical model, Jakobsson et al. [8] estimate that the transition to a fully oxygenated Arctic Ocean happened when FS reached a width of around 40–50 km. Subsequently, Thompson et al. [6] achieve ventilation of the Arctic with a prescribed FS sill depth of 1000 m and a width of 100 km and a two-layer ocean stratification in the early Miocene. The exchange flow through FS shows an outflow of low-salinity water from the Arctic Ocean confined to a thin upper layer and an inflow from the saline Atlantic Ocean below. However, the Thompson et al.’s [6] model did not consider oceanic temperature variations and wind-forcing, and their calculations are based on a two-layer stratification only. Today, there is a fairly complicated three-layer stratification in the Arctic Ocean with a cold bottom layer [14, 15] and partly wind-driven circulation.

In our study, we provide an enhanced simulation with advanced oceanic and atmospheric boundary conditions at the Miocene/Oligocene boundary. We like to understand how fast the oceanographic changes from a two-layer to present-day conditions are possible. Furthermore, we apply a climate model to assess the sensitivity of stratification and circulation in the Arctic Ocean to the opening of FS and different levels of atmospheric CO2 concentrations. The model simulation applies a Miocene-bathymetric reconstruction [10], which is believed to capture the major bathymetric features, such as different sub-basins and sills. Furthermore, CO2 changes are used relative to early–middle Miocene (about 23–15 Ma) boundary conditions. In our numerical experiments, we investigate several different opening geometries for the FS to constrain the regional and global climate impacts of this tectonic event.

Materials and methods

We use the Earth System Model COSMOS, which incorporates the atmosphere model ECHAM5 [16], the ocean model MPI-OM [17], and the land-vegetation model JSBACH [18]. The atmosphere model is used at T31 (~ 3.75° × 3.75°) horizontal resolution with 19 vertical layers. The ocean model has a resolution of GR30 (3° × 1.8°) and has 40 vertical layers. The spatial resolution increases to ~ 30 km towards the grid poles at Antarctica and Greenland. Such a better resolution close to the grid poles enhances the representation of physical processes for the deep-water formation in Nordic, Labrador, and Weddell seas. The interactive exchange of fluxes and energy between ocean and atmosphere is handled via the coupler OASIS3 [19]. Our model setup has already been used to analyse the Miocene warm climate [3, 7, 20,21,22], the Pliocene [23], the Last Glacial Maximum [24, 25], the Holocene [26, 27], and the last millennium [28]. This setup is identical to the study of Hossain et al. [7] describing the early-to-middle Miocene time period (~ 23–15 Ma).

Hossain et al. [7] investigated the impact of single ocean gateway, FS, and Greenland–Scotland Ridge (GSR), respectively, on the Arctic Ocean circulation and the combined effect of both. They found a non-linear impact of gateway depth on the water mass exchange and ocean circulation that is mainly driven by the effect of gateways subsidence and their interaction. In this study, we investigate if and how different FS widths could have affected the global ocean circulation and climate between 23 and 15 Ma. Our primary focus is on the sensitivity of Arctic Ocean stratification and circulation controlled by FS gateway configurations and CO2 changes.

In our reference simulation (MIO_FW500), we prescribe an atmospheric CO2 concentration of 450 ppm, with an FS width of ~ 500 km and fixed gateway depths of ~ 1500 mbsl. In the Arctic Ocean, the FS represents the only early Miocene gateway of the Arctic Ocean, since other shallow connections to the world’s ocean like the Barents Sea, Canadian Archipelago, and Bering Strait evolved only after the middle Miocene [29, 30]. Furthermore, the Panama Seaway is not yet closed and connects the Pacific and Atlantic oceans. Also remnants of the Tethys exist [4]. We integrate our model for 4 kyrs to minimize salinity/temperature trends in the deep ocean after initialisation with the present-day conditions. Sensitivity experiments are performed for the FS by varying its width (Table 1). After model integration of 2 kyrs, the final 100 yrs of model simulation are used for analysis.

Table 1 List of sensitivity experiments including relevant model parameters

Starting from a narrow FS width of ~ 50 km, in a set of six model simulations, we gradually widened the ocean gateway to 105, 222, 286, 352, and 500 km, respectively (Table 1). As deepening of the FS between ∼2000 and 1500 mbsl might be important for the deep-water exchange [9], the sill depth of FS is always fixed to 1500 mbsl. For minimum FS widths of ~ 50 km, in principle, the ocean gateway is likely wide enough to allow rotationally controlled exchange flows [31], as described by the Rossby radius of deformation. However, in the model experiments MIO_FW50 and MIO_FW105 with a FS width of ~ 50 km and ~ 105 km, its geometry is only represented by one and two zonal grid boxes, respectively, owing to spatial limits in the resolution of the ocean model component. Apart from the FS gateway sensitivity experiments, we performed additional simulations at different levels of atmospheric CO2 (280, 450, 600, and 840 ppm; Table 1) that are within a broad range of reported CO2 levels representative for the Eocene–Miocene time period [32,33,34].


Ventilation of the Arctic Ocean

For minimum FS widths of ~ 50 km, our simulation indicates a northward penetration of dense Atlantic waters across a narrow gateway. It establishes a hydraulically controlled outflow of relatively fresh water in the upper layer (Supplementary Figure 1), and a deep inflow of saline and warm Atlantic water. As a result of saltwater exchange across the FS and net Arctic freshwater input (river runoff and net precipitation) via the atmospheric hydrological cycle, a vertical Arctic halocline and salinity gradient establish. The formation of vertical and horizontal salinity gradients strengthens Arctic gyre circulation following isolines of salinity and causes poorly oxygenated conditions. An inflow of salty North Atlantic waters across the FS perturbs the Arctic stratification as reflected by excursions of the characteristic halocline (Fig. 1). The vertical separation of Atlantic water inflow with respect to the mixed layer above tends to reduce the amplitude of the halocline and the baroclinic–geostrophic balance of the gyre circulation.

Fig. 1
figure 1

Impact of gateway width on vertical salinity characteristics in the Arctic Ocean. a Mean salinity profiles (‰) and b haloclines (dS/dz; ‰/m) of the Arctic Ocean for different FS gateway widths

By widening the FS from ~ 50 to ~ 105 km at fixed gateway depths of ~ 1500 mbsl, an unrestricted inflow of Atlantic water to the Arctic is possible and indicated by a prominent unperturbed Arctic halocline (Fig. 1). We detect the transition towards a bi-directional gateway circulation and ventilation of the Arctic Ocean. It establishes a three-layer exchange flow through the FS that is characterized by vertical differentiation of water masses. The outflow of relatively thin, cold, and low-salinity Arctic water is situated above a compensational inflow of warm and salty North Atlantic water and a cold bottom outflow (Fig. 2). Due to the enhanced import of saline and oxygen‐rich Atlantic water through the FS, the Arctic subsurface waters eventually became saltier and oxygenated [6, 8]. With the establishment of a bi-directional circulation regime and an Arctic halocline (Fig. 1), the through flow into the Arctic Basin causes the reorganization towards a ventilated Arctic salinity regime.

Fig. 2
figure 2

A cross-section of the Arctic Ocean annual mean temperature (in K) for the model experiments: a MIO_FW50, b MIO_FW105, c MIO_FW222, d MIO_FW286, and e MIO_FW352

Widening of the FS towards ~ 286 km or above further strengthens a more effective cross-sectional water mass transport (Fig. 3 and Supplementary Figure 2). The strait becomes wide enough that the effect of the Earth’s rotation alters the water flow in the upper layer through the FS to a rotationally controlled bi-directional flow, rather than hydraulically controlled and the establishment towards a modern prototype exchange flow. It is characterized by the horizontal differentiation between the southward directed outflow of Arctic Basin at the western continental slope of the strait and the northward directed Atlantic inflow to the East. Although a modern-like wide FS gateway configuration allows unrestricted ocean water interchange and thus reducing the Arctic halocline, we still obtain stronger than PI vertical salinity contrasts. This is mainly because of a relatively fresh Arctic surface layer fed by river runoff and net precipitation balanced by salty southern sourced Atlantic water. The total outflow of low-salinity surface water via the FS is greater than the total inflow (Supplementary Table 1). Much of this exchange reflects recirculation within the strait, although parts of it enter the Arctic Ocean and contribute to ventilate deep waters.

Fig. 3
figure 3

Ocean velocity at the surface (6 m water depth, m/s) for the experiments of: a MIO_FW50, b MIO_FW105, c MIO_FW222, d MIO_FW286, e MIO_FW352, and f MIO_FW500

Finally, we calculate the turnover time, which is defined as the ratio of total water volume in the Arctic Ocean to the total outflow through FS [6]. In our study, the total volume of water in the Arctic is 1.6293e+16 m3 and the total outflow through FS is 1.01e+6 m3/s. This results into a similar turnover time of ~ 512 yrs, compared to 480 yrs estimated by Thompson et al. [6] taken the uncertainties into account.

Stratification in the Arctic Ocean

In the model experiments MIO_FW50 at a FS width of ~ 50 km, we detect that the simulated stratification in the Arctic Ocean has a two-layer structure only, with a surface layer of shallow cold and low-salinity water situated above a deep and weakly stratified lower layer (Fig. 2). The lower layer consists of warmer saline water of Atlantic origin that is advected northward through FS and progressively mixed with the cold low-salinity upper ocean water. The exchange flow through FS is characterized by vertical separation of water masses, as shown by an outflow of relatively fresh and cold Arctic waters at the surface and a compensational inflow of warm and salty Atlantic waters below. This two-layer stratification of the Arctic Basin is caused by the net freshwater input and the reduced inflow of saline Atlantic water, which deters the possibility of forming a less saline but cold bottom layer.

The two-layer stratification in the Arctic Ocean changes with the widening of the FS and the increasing ocean water mass interchange. In the model experiments MIO_FW105, MIO_FW222, MIO_FW286, MIO_FW352, and MIO_FW500 at the FS width of ~ 105,  ~ 222,  ~ 286,  ~ 352, and  ~ 500 km (Figs. 2 and 4b), respectively, and with a sill depth of ~ 1500 m, a three-layer structure for the Arctic Ocean is established, with a cold bottom layer encountered below the warmer intermediate Atlantic layer.

Fig. 4
figure 4

A cross-section of the Arctic Ocean annual mean temperature (in K) for the model experiments: a MIO_280, b MIO_FW500, c MIO_600, and d MIO_840

Effect of CO2 concentrations on Arctic Ocean stratification

To test the sensitivity of Arctic Ocean stratification to atmospheric changes, we perform additional simulations at different levels of atmospheric CO2 concentrations (280–840 ppm), capturing a wide range of greenhouse gas variations representative for the Eocene–Miocene time period [32,33,34]. We choose only a single FS width of ~ 500 km and vary the atmospheric CO2 (280, 450, 600, and 840 ppm) accordingly. The CO2 concentration at 450 ppm reflects the standard Miocene configuration that has also been applied for the earlier presented FS gateway sensitivity experiments. The CO2 level of 280 ppm is typical for pre-industrial (PI) times. Today, ~ 415 ppm has been reached.

With the increasing of CO2 levels, we observe a warming (3–6 °C) (Fig. 4) in combination with reduced annual mean sea-ice cover in the Arctic Ocean (Supplementary Figure 3). Analyses of gravity cores [35] show that the central Arctic Ocean was more or less ice-free during middle-to-late Miocene summers for 600 and 840 ppm CO2 levels, whereas sea ice still existed during summers for 278 and 450 ppm CO2 simulations. We find that the simulated stratification in the Arctic Ocean regime for the model experiments MIO_280 has a bi-layer structure, with a cold shallow upper layer above a slightly warmer deep-water mass likely of Atlantic origin that extends down to the bottom (Fig. 4). With the increasing of CO2 levels to 450 ppm or above (MIO_FW500, MIO_600, and MIO_840), we find three-layer stratification in the Arctic Ocean, including a cold low-saline bottom layer. This three-layer structure becomes even more pronounced by increasing the CO2 level, and is most pronounced at CO2 levels of 840 ppm (Fig. 4).

Elevated atmospheric CO2 concentrations enhance the Arctic freshwater budget (Supplementary Table 1). The additional freshwater from the Arctic region is transferred into the Atlantic Ocean, combined with enhanced high latitude warming that reduces deep-water formation. This leads to a slowdown of the Atlantic Meridional Overturning Circulation (AMOC) (Supplementary Table 1). At the FS gateway, particularly, the additional Arctic freshwater export linked with the attenuated salt import of northward directed Atlantic waters decreases the baroclinity and overall salinity in the Arctic Ocean (Supplementary Fig. 4 and Supplementary Table 1).

Effect of FS widening on the climate

An isolated widening of the FS gateway from ~ 50 to ~ 105 km at fixed gateway depths of ~ 1500 mbsl provides an enhanced inflow of warmer and saltier Atlantic water to the Arctic Ocean and unrestricted water exchange (Supplementary Figure 2a, b). As a consequence, we detect mild warming (up to + 1 K), but unchanged salinity conditions in the Nordic Seas (Figs. 5, 6). Arctic temperatures also remain unaltered, but a salinity (up to + 3 psu) increase in combination with reduced sea-ice cover is detected (Supplementary Fig. 5).

Fig. 5
figure 5

The effect of a ΔFW105-50, b ΔFW222-50, c ΔFW286-50, d ΔFW352-50, and e ΔFW500-50 on sea-surface temperature anomalies (SST; in K)

Fig. 6
figure 6

The effect of a ΔFW105-50, b ΔFW222-50, c ΔFW286-50, d ΔFW352-50, and e ΔFW500-50 on sea-surface salinity anomalies (SSS; in psu)

Progressive widening of the FS gateway from ~ 50 to ~ 500 km (Table 1) shows the similar basic characteristics in temperature and salinity as widening from ~ 50 to ~ 105 km with stronger magnitudes of change for a wider FS gateway (Figs. 5, 6). The SSS changes in the Arctic appear to steadily become stronger with the widening of FS gateway. It is caused by the increased inflow of saline Atlantic water through FS which progressively mixed with the low-salinity upper ocean water.


The continuous opening of FS is well constrained by magnetic and seismic data [9] that show oceanic crust in the FS might have been formed after ~ 24 Ma [9]. Thus, during this initial opening phase, a shallow water exchange between the Arctic Ocean and the North Atlantic [36, 37] is likely. Both plate kinematics models and geophysical data [9, 38] indicate that the opening of FS started earlier than suggested by previous paleo-oceanographic studies [39] based on micropaleontology and biostratigraphy. With time, the FS widened as a result of a large strike-slip movement between Svalbard and Greenland. A paleo-bathymetric model based on magnetic and regional seismic data [90] shows that the first deep-water exchange between the Arctic Basin and the Nordic Seas was likely between ~ 17 and 20 Ma. According to this model, the FS deepened to sill depths > 1500 mbsl at ~ 17 Ma. A paleo-bathymetric reconstruction indicates the transition to a fully ventilated Arctic Ocean at around 17.5 Ma allowing an unrestricted exchange of warm and saline Atlantic water and colder Arctic water [8, 40]. The ventilation was achieved in a relatively short period, since the drill cores show an abrupt change to modern sedimentation at that time instant. According to the age–width estimation using geophysical data and a recent paleo-bathymetric model [41] yields similar results. The FS began to deepen in the early Miocene and reached modern depths (> 2000 m) by the middle Miocene (13.7 Ma [8]).

The simulated climate shows a global warming that matches the global mean temperature reconstruction [42] suggesting warming (+ 6 K) with respect to PI conditions [3]. The Arctic Ocean was also relatively warm with temperatures of about 8.4–9.2 °C (Supplementary Table 1). This is directly supported by the early Miocene (~ 18.2 Ma) sediments recovered from the Lomonosov Ridge (IODP Exp 302) during ACEX indicate SST of ~ 10 °C [43]. This also agrees with oxygen isotope analyses of fishbone carbonate [44] in the ACEX cores provided a salinity of about 35 psu and temperatures of about 6 °C for the early Miocene. Stein et al. [45] found alkenone-derived SST of ~ 8.9 °C in the central Arctic Ocean during late Miocene (~ 7.5 Ma; Supplementary Table 2). Observational data based on a combination of satellite and direct measurements [46, 47] show a warming (~ 0.5–1.5 °C) in the Nordic Seas with a CO2 rise from ~ 300 to 400 ppm between 1901 and 2015 [48, 49]. In our Miocene simulation with increasing CO2 levels from 280 to 450 ppm, we observe warmer (~ 2.5–4.5 °C) Nordic Seas (Supplementary Figure 6). In general, our simulations could reproduce a warmer than the present-day Arctic Ocean during the Miocene [2, 50] with a stronger hydrological cycle (Supplementary Table 1 [51]).

The geometrical widening of the FS could have been important for climate and global ocean circulation by influencing the production of NADW, and initiating and intensifying the AMOC [Supplementary Figure 7; 8, 11, 30, 52]. The transition from poorly oxygenated towards fully ventilated (oxygenated) Arctic Ocean, characterized by a warmer ocean with higher salinities, as suggested by sediment records from the Lomonosov Ridge (ACEX core analyses), has been attributed to the opening of FS [8, 40, 44].

In our numerical simulations, we have achieved a ventilated Arctic Ocean with a width of ~ 105 km. This agrees with the findings of previous studies [6, 53], although they have applied a shallower FS sill depth (1000 m) than used in our study (~ 1500 m) to induce ventilation. Hence, we do not expect a big difference to Thompson et al. [6, 53] from the point of ventilation changes if we would apply a similar depth. Furthermore, we used a fully coupled Earth System Model including feedbacks in the atmosphere–ocean–land system. Hence, the similarity of the dynamics of Arctic ventilation in response to FS widening in Thompson et al. [6, 53] and our study highlights the importance of an oceanic control. A narrowing of the FS in the late Miocene by the Hovgård Ridge [11] might not have had a large impact as long as the remaining width was more than 100 km. The ventilation of the Arctic Ocean is established by inflow through FS of North Atlantic originated deep-water and injection of brine-enriched dense shelf waters [14, 15]. We estimate a turnover time of ~ 512 yrs in our study which is similar to the time estimated in Thompson et al. [6].

Our model experiments with an FS width at around 105 km or more show a three-layer structure of Arctic stratification, with a colder shallow upper layer above a warmer intermediate layer of Atlantic origin and a relatively less-warm bottom layer. The present-day Arctic Ocean also has a three-layer stratification, which depends crucially on temperature variations, the sea-ice formation on the shelves, and wind-forcing [14, 15]. The exchange flow through FS shows a three-layer structure with a thin upper layer of low-salinity cold outflow from the Arctic Ocean, an intermediate inflow of saline and warmer water from the Atlantic Ocean below, and a cold bottom layer of Arctic origin. However, it is in contrast to the findings of previous studies [6, 54] which found a two-layer stratification using a Miocene bathymetry.

At the wider FS gateway, warmer Atlantic water flushes the intermediate depths and progressively mixed with the cold low-salinity upper ocean water and bottom layer. It causes a gradual non-linear salinity increase (up to + 2 psu) and warming (up to + 1 K) in the Arctic Ocean. The flow of warm waters from the North Atlantic enters the Nordic Seas and Arctic Basin through FS, mostly between 200 m depth and the mixed layer (Figs. 2, 4), leading to substantial warming in the Arctic Ocean [54]. This non-linear salinization process and warming control the stratification.

Elevated atmospheric CO2 enhances the Arctic freshwater budget, warming (Supplementary Table 1), and a more accentuated halocline [3]. An intensified stratification is established by an enhanced atmospheric hydrological cycle with increased river runoff and net precipitation (precipitation minus evaporation) in the Arctic Ocean (Supplementary Table 1; [40]) that tends to freshen the relative fresh Arctic surface waters [3]. The lack of relatively saline inflow through the FS might also contribute to Arctic sea-surface freshening during the Miocene [54]. Below the Arctic freshwater layer, there are the southern sourced North Atlantic waters, which cause an increase in salinity and temperature due to high evaporation in the low latitudes.

In the northern hemisphere, simulated annual sea-ice concentrations are significantly higher for large parts of the Arctic when FS was narrow and CO2 levels low (Supplementary Figures 3, 8), which was due to cooler mixed-layer temperatures in combination with weaker NADW formation [54]. Conversely, sea-ice decreases for a wider FS gateway and higher CO2 levels. A perennial Arctic sea-ice cover likely occurred from the middle Miocene onwards [40, 55,56,57]. These data are in contrast to geological analyses [35] that show ice-free warmer summer conditions in the Arctic Ocean during the middle-to-early late Miocene. Sea-ice proxy records suggest that Arctic sea-ice cover was modest between 17.5 and 16 Ma and more prevalent following and prior to this time period [58]. There is likely excessive Arctic sea-ice in our Miocene simulation [54].

The relatively short ventilation time is supported by the abrupt changes in sediment composition in the ACEX core analyses [8]. Although FS is fairly shallow and narrow in our Miocene simulations, the resulting ventilation timescales seem to be shorter than or comparable to corresponding estimates based on chemical tracer studies [59, 60] in the present-day Arctic Ocean. In general, the current model reproduces geological observations on the climate conditions of the Arctic Ocean. Our model simulations do not show dramatic changes in the global circulation pattern and global climate questioning the role of the FS in triggering/enabling the Northern Hemisphere Glaciations (NHG) during early-to-middle Miocene. Several studies have proposed that the closure of the Panama Seaway [61, 62], a decline in atmospheric CO2 [63,64,65], and tectonic uplift of plateaus and mountains in the high northern latitudes [11] are the important factors for the onset and intensification of the NHG during the Pliocene around 2.7 Ma [11, 66,67,68].


By means of an Earth System Model, we have qualitatively analysed the impact of FS opening in controlling ventilation of the Arctic Ocean during early-to-middle Miocene in different model experiments. We explore the role of the FS width and CO2 concentrations in the establishment of the modern-like stratification in the Arctic Ocean and exchange flow through FS. Our simulations show that a progressive widening of the FS causes unaltered salinity conditions and a mild warming in the Nordic Seas. Arctic water temperatures remain unaltered and salinity changes appear to steadily become stronger. For a sill depth at around 1500 m, we have achieved ventilation of the Arctic Ocean regime with an FS width of ~ 105 km. At this depth and width at around 105 km or more, we observe a modern-like three-layer stratification in the Arctic Ocean, with a shallow surface layer of cold and low-salinity water situated above a deep and warmer layer of Atlantic origin and a cold bottom layer. The exchange flow through FS also shows a three-layer structure. The ventilation mechanisms and stratification in the Arctic Ocean simulated in our study using a significantly shallower and narrower FS during early Miocene are comparable to the present-day ocean basin. In general, our model reproduces geological observations on the climate conditions of the Arctic Ocean. However, the simulations do not show dramatic changes in the global circulation pattern and global climate questioning the role of the FS in triggering/enabling the NHG.

In future studies, the presented framework might be a good tool to interpret high-resolution sediment records and data from up-coming drilling projects that target the past climate evolution in the North Atlantic–Arctic sector.