The expectation of spatially coherent SST anomalies between the North Atlantic and the Norwegian and Iceland seas is rooted in the observational period and investigations of SST anomalies at specific stations along the pathway of the North Atlantic Current (Årthun and Eldevik, 2016; Årthun et al., 2017). A comparable analysis of the observational record is done here to confirm if the expectation of spatial coherence holds when looking at averages over larger domains encompassing the North Atlantic, Norwegian and Iceland seas. The data are averaged over three box domains, as shown in Fig. 1, to represent the three sites in the Pliocene reconstructions (Sect. 2.3). The same domains are used in the CMIP6 model analysis (Sect. 2.2). We consider it to be that SST averaged over the domains better represents the variability than a single grid point. The domains are chosen as follows: to represent the site in the NE North Atlantic, we use a domain covering the northeastern part of the Subpolar North Atlantic (49–57∘ N, 35–14∘ W); to represent the site along the NwAC (North Atlantic Current), we use a box over the eastern Norwegian Sea (62.5–73∘ N, 0–16∘ E); and finally, to represent the site in the Iceland Sea, we use a box covering the major part of the Iceland Sea (66–72∘ N, 18–10∘ W). The analysed dataset is from the Met Office Hadley Centre (version 1.1) and provides monthly global SST on a 1∘ latitude–longitude grid over the period from 1870 to 2012. A detailed description of the dataset is given in (Rayner et al., 2003). In this study we use the annual mean SST to document the existing SST anomalies and the spatial relation of these between the North Atlantic and the Norwegian and Iceland seas. To investigate multidecadal timescales over the HadlSST dataset, we apply a 5-year running mean. SST anomalies are calculated relative to the mean of the respective records (1870–2012).

The current generation of global climate models is available through CMIP6. CMIP6 provides a range of climate change experiments to the end of this century and beyond. To address whether or not the spatially non-coherent SST response may also occur in the future under less-extreme atmospheric-CO2-forcing scenarios than for RCP8.5, we use monthly gridded SST data from the SSP126 experiment covering the time period from 2021 to 2100 with an approximate radiative forcing of 2.6 W m-2 and a relatively low level of global warming (it is called the 2 ∘C scenario) by 2100. CO2 concentrations reach 445 ppm by 2100 (Meinshausen et al., 2020), which is at the high end of the Pliocene CO2 range (Bartoli et al., 2011; de la Vega et al., 2020). Similarly to how we treat the HadlSST data, we assess the annual mean SST from each of the model simulations. A 5-year running mean has been used to smooth the interannual variability. Because of the 5-year-running-mean filter, the time series are shown for the period 2023 to 2098. SST anomalies are calculated relative to the mean of the respective records (2021–2100).

The CMIP6 archive offers model output from many models. In this study, we have chosen to analyse three different models that have different equilibrium climate sensitivity (ECS; Meehl et al., 2020; Seland et al., 2020), leading to different amounts of warming by 2100: CNRM-ESM2-1, which has the highest sensitivity of the three models (ECS = 4.8); NorESM2-MM, which has the lowest sensitivity (ECS = 2.5); and MPI-ESM1-2-LR, which is in between in terms of sensitivity (ECS = 3). In the analysis herein, we use one member from each model (i.e. one simulation from each of the three selected models). Some additional model simulations have been included to check the responses to a more aggressive warming scenario (SSP585 experiment; NorESM2-MM), a different resolution in the atmosphere (1∘ versus 2∘; NorESM2-MM vs. NorESM2-LM), and more members (10 members compared to one single member; MPI-ESM1-2-LR). The model data are averaged over the same domains as analysed for the observational data, namely the NE North Atlantic (49–57∘ N, 35–14∘ W), the Norwegian Sea (62.5–73∘ N, 0–16∘ E), and the Iceland Sea (66–72∘ N, 18–10∘ W). SST anomalies relative to the mean of the 5-year-running-mean-filtered time series are calculated for CNRM-ESM-2-1, one of the 10 MPI-ESM1-2-LR members, and for NorESM2-MM (SSP126) and NorESM2-MM (SSP585). We focus on the SST anomalies for the three domains and the relation between these at the end of the century (2068–2098).

To see if spatially non-coherent SST anomalies also occur between the North Atlantic and the Norwegian and Iceland seas under a past warm-climate period in equilibrium with atmospheric CO2 concentrations comparable to the SSP126 experiments, we use a compilation of previously published Pliocene alkenone U37K′ SST data from three sites from the northern NE North Atlantic (ODP Site 982; 57.5167∘ N, 15.8667∘ W; 1134.2 m water depth) (Herbert et al., 2016; Lawrence et al., 2009), the Norwegian Sea (ODP Site 642; 67.255∘ N, 2.928333∘ E; 1280.9 m water depth) (Bachem et al., 2016, 2017), and the Iceland Sea (ODP Site 907; 69.24815∘ N, 12.69∘ W; 1801.5 m water depth) (Herbert et al., 2016) (Fig. 1). Each dataset covers the time interval between 5.23 and 3.13 Ma.

The U37K′ index records the relative abundance of specific lipids (alkenones) synthesized by selected unicellular haptophyte algae living at or near the sea surface (e.g. Marlowe et al., 1984). Through the study of cultures, water samples, and surface sediments, it has been shown that the U37K′ index changes with temperature (Prahl and Wakeham, 1987; Müller et al., 1998; Conte et al., 2006; Tierney and Tingley, 2018). All records are presented here as previously published (Herbert et al., 2016; Lawrence et al., 2009; Bachem et al., 2016, 2017), using established age models and the Müller et al. (1998) U37K′–SST calibration. The near-global and linear relationship between surface sediment U37K′ values and mean annual SSTs (Müller et al., 1998) aligns closely to a culture study (Prahl and Wakeham, 1987) and has been used to calibrate and reconstruct mean annual SSTs. The standard error of estimation using this calibration is ±1.5 ∘C (Müller et al., 1998). As a biological temperature proxy, it is important to consider both the environmental and biological influences over this U37K′–SST relationship. Marked local or regional differences in the timing of alkenone production and flux to the seafloor may impart a seasonal bias to the sedimentary record (e.g. Rosell-Mele and Prahl, 2013). In a recent expansion and Bayesian analysis of the global surface sediment calibration, a stronger correlation to August–October SSTs was identified in the North Atlantic (Tierney and Tingley, 2018), i.e. in the region of our study, which may be supported by the overlap between reconstructed SSTs and autumn multi-model means for ODP Site 982 and ODP Site 642 during the KM5c interglacial at 3.205 Ma (McClymont et al., 2020). In the Nordic Seas, low salinity or high sea ice have been linked to elevated production of the C37:4 alkenone (e.g. Bendle and Rosell-Melé, 2004; Wang et al., 2021). However, this alkenone is not included in the U37K′ index and was not recorded at values of concern at ODP Site 642 (Bachem et al., 2017).

The sampling resolution of the original records varies; for Site 907 (Iceland Sea), the mean original temporal resolution was ca. 2600 years; however, from 3.33 to 3.16 Ma, the spacing between measurements ranges from 4000 to 70 000 years; for Site 642 (Norwegian Sea), the mean resolution was ca. 2500 years between 3.13 and 3.49 Ma and ca. 7200 years between 3.49 and 5.23 Ma; and for Site 982, it was ca. 2100 years between 3.13 and 4.03 Ma and ca. 40 800 years between 4.03 and 5.23 Ma. To enable direct comparison between sites, independent of differences in temporal resolution and absolute ages for the raw data points, each dataset has been resampled every 100 kyr between 5.23 and 3.13 Ma using a linear integration function in AnalySeries (Paillard et al., 1996). SST anomalies are calculated relative to the mean of the respective resampled records (5.23–3.13 Ma). The 100 kyr resampling interval is chosen to put focus on the long-term trends of each record and the background climate state upon which the shorter-term orbital variability is superimposed (Fig. 2). The shorter-term orbital variability is not well-enough resolved by all records throughout the investigated time interval to allow for a higher-resolution resampling. Hence, given the timescales considered here, the SST anomaly relations are unlikely to be orbitally forced. Furthermore, focusing on the mean state of longer intervals (the shortest time interval is 200 000 years) rather than point-to-point comparison also minimizes the impact of uncertainties from age models. Pliocene chronologies are mostly constrained by tuning to LR04 (Lisiecki and Raymo, 2005) and/or using tie points from magnetic reversals. The tuning error is generally considered to be no more than a few thousand years but may exceed 10 kyr prior to 4.3 Ma due to less-certain obliquity variance (Lisiecki and Raymo, 2005). At the timescales considered here, such errors are acceptable.

We use an idealized-topography configuration of the MITgcm (Marshall et al., 1997) to investigate the SST relationships in the study area under a range of buoyancy forcings (Fig. 1). The set-up (Fig. 3) is a Nordic-Seas-like basin separated from a truncated North-Atlantic-like source water region by a 1000 m deep ridge. Both basins are flat-bottomed, with 2000 m depth, and are surrounded by sloping sides. The model domain is closed. The boundary conditions and prescribed forcings are the following: there is a restoring boundary condition in the south maintaining the reservoir of Atlantic source water. The temperature is restored to a temperature of TA= 6 ∘C and a salinity of 35 psu, with a restoring strength of 40 W m-2 and a timescale of 1 month. In addition, SSTs are restored toward atmospheric temperatures (SATs) through the surface heat flux which is parameterized as Q=(SST-SAT)⋅G, where G= 40 W m-2 C-1. There is no interactive atmosphere. There is a constant, uniform freshwater input in the form of precipitation north of the ridge. Mechanical forcing is provided by a constant-in-time prescribed wind field (W) with westerlies over the North Atlantic and easterlies over the Nordic Seas field. The latitude of zero wind stress curl is in the middle of the Atlantic region, with cyclonic wind stress to the north and anti-cyclonic wind stress to the south.

The horizontal resolution is 10 km, and there are 30 vertical layers; the upper 20 layers are 50 m thick, and the deepest layers are 100 m thick. Water density is calculated using the formula from Jackett and McDougall (1995), with a constant Coriolis parameter, f=1.2×10-4 s-1, and vertical diffusivity and viscosity of 1×10-5 m2 s-1. Convection is parameterized with implicit vertical diffusion; the diffusivity increases to 1000 m2 s-1 for statically unstable conditions. Horizontal viscosity is parameterized using the Smagorinsky closure (Smagorinsky, 1963); typical values are 30 m2 s-1 for the boundary current region. Temperature and salinity are advected using a third-order flux-limiting scheme.

Since one of the key drivers for inflow of Atlantic Water to the Norwegian Sea is buoyancy forcing and the production of dense overflow water that must be compensated for (Furevik et al., 2007), we change the SAT (G) and the freshwater (in the form of precipitation) (P) north of the ridge to study the impact of buoyancy forcing on the relationships between SSTs in the North Atlantic, Norwegian Sea, and Iceland Sea. The SAT and freshwater are changed as shown in Fig. 4 and Table 1. Note that the SAT over the restoring region is the same for all experiments. The idealized model is run for 30 years to near steady state, and we present results from the last 5 years of the runs, which are compared to the results from the relevant reference experiment. We are therefore not studying transient changes but differences between equilibrium states.

Buoyancy changes are forced north of the ridge due to the nature of the model set-up. The restoring boundary conditions in the south are also a forcing, representing both the (infinite) source of Atlantic Water and the experiment's energy source (heat and buoyancy input). As the surface forcing is applied north of the ridge, water mass transformation takes place, and a consistent ocean circulation is set up, including the setting of the hydrography of the different regions. The southern boundary energy input and northern surface heat loss balance when the model has reached (quasi)-equilibrium. The northern and southern regions are accordingly equally important for the experiments.

We want to investigate the responses to changes in buoyancy caused by either an SAT or a freshwater change. In addition, we want to see if the initial state of the ocean impacts the response to a SAT change, specifically testing if the response differs if we start out from the fresher Nordic Seas. Therefore, we define three reference experiments, namely REF-1 (G0 and P1), REF-2 (G1 and P1), and REF-3 (G0 and P3) (Table 1). REF-1 is set up to investigate the oceanographic responses to a gradually decreasing SAT gradient between the North Atlantic and the Nordic Seas under constant freshwater forcing by increasing the SATs over the Nordic Seas. For the REF-2 experiments, the buoyancy is changed by gradually increasing the freshwater over the Nordic Seas, while SAT is kept constant. The REF-3 experiments are similar to the REF-1 experiments in the sense that SAT over the Nordic Seas is increased; however, the initial state of the Nordic Seas is fresher for REF-3 than for REF-1. Hence the REF-3 experiments are set up to see how the initial state of the ocean may impact the responses to increased SAT over the Nordic Seas.

The combination of the prescribed wind stress and the steep yet sloping coastal boundary supports a cyclonic boundary-intensified circulation around the Nordic Seas. The reference experiments have an ocean circulation which represents the main characteristics of the North Atlantic (south of the ridge), with an anticyclonic gyre in the subpolar latitudes and a cyclonic gyre further south. The buoyancy forcing from the prescribed surface temperature and salinity results in a gradual meridional temperature decrease and a similar salinity decrease, mimicking northern heat loss and freshwater input (Fig. 5). The thermal forcing dominates, and there is net northern buoyancy loss. There is warm and saline inflow to the Nordic-Seas-like basin and a colder and fresher outflow. The densest and coldest water is found in the relatively motionless and weakly stratified interior of the Nordic Seas. The overall temperature contrast between the dense-water interior and the Atlantic source water region reflects the temperature range of the prescribed surface air temperature. Waters are continuously exchanged between the buoyant boundary current and the interior by lateral eddy mixing. Heat is thus not only lost from the boundary current by air–sea interaction but also by lateral heat loss to the interior, where it is also given up to the atmosphere.

For the MITgcm, the Norwegian Sea domain is defined as a box in the eastern boundary current region, while the Iceland Sea domain is represented by the interior ocean north of the ridge (Fig. 5). The definition of these domains, as the domains used for the observations and CMIP6 results (Sect. 2.1; Fig. 1), is directed by the location of the Pliocene sites, representing the Norwegian boundary current and the interior Iceland Sea.

The definition of the North Atlantic domain is somewhat different; it is, for simplicity, defined as the North Atlantic restoring region (Fig. 5), restored to 6 ∘C for all experiments (Fig. 4a). With this restoring, the state of the North Atlantic source water that eventually becomes the inflow to the Nordic Seas is essentially known. Also, the model North Atlantic is much less directly impacted by the prescribed changes in buoyancy forcings than the Nordic Seas (Fig. 4) and is also less impacted in consequence, as evident in Fig. 9. Directly related, it is the relative temperature (density) difference of the model ocean that constrains the flow and thus the results. The nonlinearity of the equation of state will only be in effect for large excursions in the absolute values of the restoring temperature and salinity between the different experiments.

In short, the MITgcm experiments assess the state of the Nordic Seas, including that of the Norwegian and Iceland seas, relative to that of the North Atlantic. The summary of experiments in Table 1 reflects this relative perspective.

We identify spatially coherent and non-coherent SST anomaly relationships between the North Atlantic and the Norwegian and Iceland seas by comparing the temperature of the sensitivity experiment with the relevant reference experiment. Change in a region is classified as an SST anomaly when the temperature change between two experiments exceeds 2σ(SSTreference_experiment), with σ calculated after temporally averaging the model SSTs. The North Atlantic is restored to constant temperatures as mentioned above. Even if it deviates (slightly) within what is allowed for by the restoring (see Sect. 2.4), it remains essentially constant and non-anomalous. Thus, as also alluded to above, change in the SST anomaly relationship between the three regions exists if there is a temperature change in either the Norwegian Sea or the Iceland Sea, or in both, larger than 2σ.

On multidecadal timescales the annual SST anomaly, as seen in the HadlSST dataset, varies between -0.8 and +0.8 ∘C (Fig. 6). As described in the introduction, the spatially non-coherent SST anomaly signal seen on shorter timescales should in theory no longer be of importance on multidecadal timescales, and we see a spatially coherent SST anomaly relationship between the North Atlantic, the Norwegian Sea, and the Iceland Sea (Fig. 6). The spatially coherent SST anomaly relationship between the North Atlantic, the Norwegian Sea, and the Iceland Sea is robust, showing positive and significant correlations between the detrended time series (both the Norwegian Sea and the Iceland Sea are correlated with the North Atlantic). The relationship also holds for different filtering of the time series, i.e. running means with a 5-year window, 10-year window, and 15-year window (not shown). On multidecadal timescales, these regions follow, to a large extent, the Atlantic multidecadal variability (AMV), with a warm phase in 1930–1970 and a cold phase in 1970–1990 (Knight et al., 2005).

The three models, CNMR-ESM2-1, MPI-ESM1-2-LR, and NorESM2-MM, show different results, both with respect to their SST climatology and the SST anomalies in each region of interest for the end of the 21st century (the last 3 decades, 2068–2098) (Fig. 7). The mean SST of the 5-year-running-mean-filtered time series (2023–2098) of the NE North Atlantic is fairly similar in the three models, but that of the Norwegian Sea differs to some extent, with MPI-ESM1-2-LR being warmest (9.5 ∘C), NorESM2-MM being coldest (6.8 ∘C), and CNRM-ESM2-1 being in between (7.8 ∘C). The mean SST in the Iceland Sea differs to a large extent among the three models, again with MPI-ESM1-2-LR being warmest (5.6 ∘C) and NorESM2-MM being coldest (0 ∘C).

CNRM-ESM2-1 shows larger SST anomalies for the Norwegian Sea and the Iceland Sea than the two other models (Fig. 7b). This is consistent with CNRM-ESM2-1 being the most sensitive, or more rapidly responding, model (as described in Sect. 2.2). Based on CNRM-ESM2-1, we find a dominantly cold SST anomaly in the North Atlantic and a warm SST anomaly in the Norwegian and the Iceland seas toward the end of the 21st century (2068–2098). Thus, for the SSP126 scenario, CNRM-ESM2-1 suggests that the North Atlantic SST anomaly will differ from the SST anomalies of the Norwegian and Iceland seas at the end of the 21st century.

Compared to the CNRM-ESM2-1 results, both MPI-ESM1-2-LR and NorESM2-MM show much smaller SST anomalies at the end of the 21st century relative to the model annual mean SST over the next century (Fig. 7b). Considering 10 different members from MPI-ESM1-2-LR, we find that the results from the individual members do not differ to a large extent (Fig. 7a); none of the members show anything but minor annual mean SST variability for any of the domains. The difference between the members is less than the amplitude of the changes in CNRM-ESM2-1. On the other hand, considering a more aggressive scenario (SSP585) for NorESM2-MM, we find a clear warm anomaly in the Norwegian and Iceland seas (Fig. 7b). A cold, but small, SST anomaly is seen for the North Atlantic. Hence, the sensitivity of the model impacts the result. Furthermore, we find that lowering the horizontal resolution in the atmosphere entails higher SSTs in the Iceland Sea at the end of the century relative to the results from the model version with a higher horizontal resolution in the atmosphere. A lower horizontal resolution in the atmosphere does not, however, have a clear effect on the SSTs of the Norwegian Sea or the North Atlantic (Fig. 7a).

The spatially incoherent SST anomaly relationship found in CNRM-ESM2-1 SSP126 and NorESM2-MM SSP585 in the last 3 decades (2068–2098) (Fig. 7) is sensitive to the chosen period and is not robust over the whole projected period (2023–2098). Correlations are overall nonsignificant between the time series from the three regions. Positive correlations are indicated for NorESM2-MM SSP585 (both the Norwegian Sea and the Iceland Sea are correlated with the North Atlantic), but only if the time series are detrended. In the latter case, the spatially incoherent SST anomaly relationship in the last 3 decades is not found in NorESM2-MM SSP585.

The relation between SST anomalies of the North Atlantic, Norwegian Sea, and Iceland Sea is not constant through the Pliocene. The following two different spatially non-coherent SST anomaly relations are documented: (1) the Norwegian Sea SST anomaly differs from the SST anomalies of the North Atlantic and the Iceland Sea, either with a warm Norwegian Sea anomaly corresponding with a cold anomaly in the North Atlantic and Iceland Sea (first yellow period in Fig. 8; 3.63–3.93 Ma) or the opposite, with a cold anomaly in the Norwegian Sea corresponding to a warm anomaly in the North Atlantic and Iceland Sea (second yellow period in Fig. 9; 4.73–4.93 Ma); (2) a cold anomaly in the North Atlantic and the Norwegian Sea corresponds to a warm anomaly in the Iceland Sea (blue period in Fig. 9; 3.23–3.43 Ma). All these Pliocene SST anomaly relationships are different from what we see in the CMIP6 future runs (warm Norwegian Sea and Iceland Sea, cold North Atlantic). In addition to the spatially non-coherent SST anomaly relations, there are three time periods during the Pliocene that show a spatially coherent SST anomaly relationship, comparable to what we find for the observation-based data (grey time periods in Fig. 8; 4.03–4.33, 4.43–4.63, and 5.03–5.23 Ma).

By changing the buoyancy forcing as shown in Figs. 4 and 5, we can produce the following three different spatial SST anomaly relationships in the idealized model: spatially coherent SST anomalies in all three regions (grey experiments, Table 1 and Fig. 9b), Norwegian Sea and Iceland Sea SST anomalies that are different from the North Atlantic (red experiments, Table 1 and Fig. 9), and an Iceland Sea SST anomaly that differ from the North Atlantic and Norwegian Sea SST anomalies (blue experiments, Table 1 and Fig. 9). Hence, these idealized experiments can capture two of the three SST anomaly relations found during the Pliocene (grey and blue time periods in Fig. 8) and the SST anomaly relationship found in the CMIP6 future runs (Fig. 7). Table 1 summarizes the experiments.

The first set of four experiments has the same freshwater forcing (P1) but a decreasing SAT gradient (i.e. increasing SAT over the Nordic Seas (G1–G4)) relative to REF-1. As SAT increases over the Nordic Seas, the SST pattern shifts from the Iceland Sea SST anomaly being different from the North Atlantic and the Norwegian Sea (G1) to the North Atlantic SST anomaly being different from the Norwegian and Iceland seas (G2–G4) (Fig. 9a).

In the next set of experiments, SAT is kept constant (G1) while the freshwater over the Nordic Seas is increased (P2–P5) relative to REF-2. With increasing freshwater over the Nordic Seas, the SST pattern shifts from one of spatial coherence (P2 and P3) to the Iceland Sea SST anomaly being different from the North Atlantic and the Norwegian Sea (P4) to the North Atlantic SST anomaly being different from the Norwegian and Iceland seas (P5) (Fig. 9b).

In the last set of experiments, the decreasing SAT gradient experiments (G1–G2) are repeated with fresher Nordic Seas (P3) relative to REF-3. As for REF-1, the SST pattern shifts from the Iceland Sea SST anomaly being different from the North Atlantic and the Norwegian Sea (G1) to the North Atlantic SST anomaly being different from the Norwegian and Iceland seas (G2) when we increase the SAT over the Nordic Seas (Fig. 9c).

In addition to identifying the SST anomalies in the MITgcm experiments, information about the density difference over the ridge, the mean inflow velocity across the sill, the mean velocity in the boundary current, the net heat transport over the sill, and the maximum overturning streamfunction at the sill is extracted for each experiment (Table 1). This information will be used in the discussion, exemplifying oceanographic responses to specific buoyancy changes as seen in the MITgcm experiments.

Reducing the SAT gradient, i.e. warming the atmosphere over the Nordic Seas, reduces the heat loss from the ocean and warms the SSTs in the Nordic Seas. Compared to the reference experiment, the following are smaller: the north–south density difference, the mean inflow velocity across the ridge, the boundary current velocity in the Nordic Seas, the net heat transport across the sill, the maximum overturning circulation across the sill, and the lateral eddy heat transport in the Nordic Seas (Table 1). The weaker ocean circulation transports less heat to the Nordic Seas. In addition, as the Norwegian Sea boundary current is slower in warmer experiments, the water of the boundary current experiences more cooling as it travels the Nordic Seas, allowing for a larger heat loss in the Norwegian Sea region.

For the experiments with a small SAT change (G1 relative to REF-1 and G1 relative to REF-3; blue experiments in Table 1 and Fig. 9a, c), the increased heat loss from the Norwegian Sea boundary current and the decreased poleward heat transport are partly able to compensate for the increased atmospheric warming in the Norwegian Sea, and thus there is no temperature change larger than 2 × SD in relation to the relevant reference experiment in the Norwegian Sea. However, this is not the case for the Iceland Sea, where the SSTs increase. We consider this result to be representative for a situation when the SST anomaly of the Iceland Sea differs from the North Atlantic and Norwegian Sea SST anomalies – hence, when there is an SST field that breaks the expectation of spatially coherent SST anomalies. The amplitude of the MITgcm Iceland Sea SST anomaly is, in this case (G1 relative to REF-1 and G1 relative to REF-3), approximately one-third of the Iceland Sea SST anomaly reconstructed for the Pliocene (3.43 to 3.23 Ma) when the SST anomaly of the Iceland Sea differs from the North Atlantic and Norwegian Sea SST anomalies.

For larger SAT warming over the northern basin (G2–G4 relative to REF-1 and G2 relative to REF3; red experiments; Table 1 and Fig. 9a, c), both the Iceland and Norwegian seas experience a temperature increase larger than 2 × SD in relation to the relevant reference experiment. There is more warming in the Iceland Sea than in the Norwegian Sea, and the difference increases with larger SAT warming (i.e. weaker SAT gradients). Thus, the absolute temperature for the Iceland Sea is more like the Norwegian Sea for a warmer atmosphere over the Nordic Seas. The larger SST change in the Iceland Sea can be explained by a combination of reduced heat transport to the Nordic Seas and a slower Nordic Seas boundary current, which allows for larger heat loss in the Norwegian Sea region and more cooling of the water as it travels the Nordic Seas. The increased heat loss and decreased poleward heat transport counteract the general atmospheric warming over the Norwegian Sea more than over the Iceland Sea. The amplitude of the SST anomalies in the Iceland and Norwegian seas, as seen in CNRM-ESM2-1 SSP126 and NorESM-MM SSP528 at the end of the century, is within the range of the respective MITgcm SST anomalies.

Increasing the north–south salinity gradient across the sill (REF-2 experiments) gives similar results; no SST change larger than 2 × SD in relation to REF-2 is found in the two regions for small changes in freshwater forcing (P2–P3 relative to REF2), an SST change is found only in the Iceland Sea for medium changes in freshwater forcing (P4 to REF-2), and SST changes are found in both the Iceland and Norwegian seas for the largest changes in freshwater forcing (P5 to REF-2, equal to a freshwater increase from 1 to 10×1e-9 m s-1 (Fig. 4)) (Table 1 and Fig. 9b). Fresher Nordic Seas weaken the north–south density gradient, which weakens the poleward heat transport and the boundary current velocity. The slower boundary current allows for more heat loss to the atmosphere in the Norwegian Sea, and the region cools. In these experiments, there is no prescribed warming of the atmosphere, so temperature change is only set by the changes in ocean circulation. The slower boundary current also reduces the eddy heat fluxes from the boundary to the interior, and hence, the Iceland Sea region also cools. As for the experiments where we increase the atmospheric temperature over the Nordic Seas, the Iceland Sea SST anomaly in the respective Pliocene case is larger than the MITgcm SST anomaly response to a medium freshwater change (P4 to REF-2). The MITgcm response to the largest freshwater forcing (P5 to REF-2) is at the lower end of the SST anomalies seen for the CMIP6 models at the end of the century.

Within the investigated parameter space, we have not found a situation where the increased heat loss and decreased poleward heat transport more than compensate for the increased atmospheric warming so that the Norwegian Sea temperature change is opposite to that of the other two regions (i.e. the situation seen for the yellow time periods in Fig. 8).

We have limited our study to the impact of buoyancy forcing on SST relationships. Using a similar model configuration, Spall (2011, 2012) report on the impact of changing sill depth on the temperature of the interior region (the inflowing temperature; i.e. Norwegian Sea temperature is assumed to equal that of the source region). A deeper sill increases the temperature of both the boundary current and the interior of the Nordic Seas. The difference in temperature between these two regions decreases for larger sill depths. The changes are associated with a strengthening of the meridional overturning circulation across the sill and, to a lesser extent, an increase in heat transport across the sill.

Spatially coherent SST anomalies in the North Atlantic and the Norwegian and Iceland seas are seen at multidecadal timescales in the HadlSST dataset (Fig. 6) and over three Pliocene time intervals covering hundreds of thousands of years, specifically 5.23–5.03, 4.65–4.43, and 4.33–4.03 Ma (Fig. 8). In the observation-based record, the spatial coherence is linked to the continuous northward propagation of heat anomalies from the subpolar region and toward the Arctic, taking about 5–10 years to propagate north, with a frequency of about 14 years (Årthun et al., 2017). Spatially coherent SST anomalies are also seen for more than half of the Pliocene time interval considered in this study, and therefore we also consider spatial coherence to be the norm for Pliocene climate, reflecting the connection of the three regions by ocean circulation (Fig. 1).

From the idealized experiments, spatial coherence is seen under weak freshwater perturbations over the Nordic Seas under constant SATs (Table 1 and Fig. 9b). The responses, however, are small, and none of the selected output parameters from the idealized experiments show responses larger than 2 × SD relative to REF-2 (Table 1). During the cold, spatially coherent interval of the Pliocene, neither the overturning circulation nor the intermediate depth ventilation of the Norwegian Sea deviated from its mean Pliocene conditions (Table 2), consistent with the weak responses seen in the MITgcm experiments for the comparable SST anomaly relation. During the warm, spatially coherent intervals of the Pliocene, the overturning circulation, as derived from the % North Atlantic Deep Water (%NADW) (Table 2), was, however, somewhat stronger than the Pliocene mean, while the intermediate depth ventilation of the Norwegian Sea was weaker than the Pliocene mean (Table 2). In line with a weaker overturning circulation during the cold relative to the warm spatially coherent intervals, a stronger North Atlantic meridional SST gradient and somewhat enhanced freshwater, or buoyancy, forcing characterized the cold relative to the warm spatially coherent SST anomalies of the Pliocene (Table 2). The characteristics associated with the warm, spatially coherent SST intervals are thus less comparable to the MITgcm results than the characteristics associated with the cold, spatially coherent SST interval.

The main difference between the Pliocene periods with warm rather than cold spatially coherent SST anomalies was a somewhat stronger freshwater influence during the cold interval, inferred from the occurrence of the sea ice marker IP25 in the Iceland Sea (Clotten et al., 2019), and that the Bering Strait was fully opened (De Schepper et al., 2015) (Table 2). While there exists evidence for sea ice in the Arctic and the Iceland Sea, it is important to note that the Arctic sea ice extent was considerably smaller than today throughout the Pliocene (Clotten et al., 2019; Knies et al., 2014). Since the Canadian Arctic Archipelago (Fig. 1) was closed throughout the Pliocene (Matthiessen et al., 2009) and since the mean Bering Strait throughflow is directed northward into the Arctic Ocean (Woodgate and Aagaard, 2005), the Fram Strait was the only Arctic Ocean exit during the Pliocene. Opening the Bering Strait allowed for inflow of Pacific water with a lower salinity to the Arctic Ocean and, consequently, enhanced the transport of freshwater from the Arctic Ocean to the Nordic Seas (Hu et al., 2015). Hence, both the increased freshwater influence through the occurrence of sea ice in the Iceland Sea and the open Bering Strait enhanced the Nordic Seas buoyancy. Sensitivity experiments performed with CCSM4 have shown that a closed Canadian Arctic Archipelago entails colder SSTs in the North Atlantic and the Norwegian and Iceland seas relative to the PlioMIP1 experiments, where both the Bering Strait and the Canadian Arctic Archipelago were open (Otto-Bliesner et al., 2016). In the idealized experiments, a weak cold, spatially coherent SST response (less than 2 × SD relative to REF-2) is found with a weak to intermediate freshwater-driven change (P3) in buoyancy. We therefore suggest that the increase in freshwater reaching the Nordic Seas from the Arctic Ocean may have caused the cold, spatially coherent SST anomaly case. We acknowledge that the source and distribution of freshwater is not directly comparable between the Pliocene and the idealized experiments; the freshwater perturbation in the MITgcm experiments is distributed equally over the Nordic Seas basin, while most of the sea ice and liquid freshwater transported from the Arctic Ocean to the Nordic Seas at any time during the Pliocene likely followed the boundary current along the east Greenland margin. The idealized model set-up includes neither the Arctic gateways nor the Arctic sea ice cover. Hence, we stress again that knowledge about changes in the Bering Strait is used to infer changes in the freshwater balance and hence buoyancy in the Nordic Seas, and similarly, the occurrence of more or less sea ice in the Iceland Sea is used to infer the likelihood that a buoyancy change took place. The data from Clotten et al. (2019) show, however, that some of this freshwater also reached the interior Iceland Sea. Less freshwater was available in the region during the Pliocene intervals with warm, spatially coherent SST anomalies, also in line with the idealized experiments where a weak warm, spatially coherent SST response (less than 2 × SD relative to REF-2) is seen for a weak salinity increase (Table 1 and Fig. 9b).

The timescales considered for the observations and the Pliocene reconstructions are very different (multidecadal versus hundreds of thousands of years); however, we consider both to reflect equilibrium, or quasi-equilibrium, responses. The timescales involved in either case are longer than the propagation-driven lag that sets up the spatially non-coherent SST anomaly relation at interannual timescales. Compared to the future scenarios which undergo transient changes due to strong CO2 forcing, we regard the era of instrumental observations to be in quasi-equilibrium. The Pliocene reconstructions represent the predominant situation over hundreds of thousands of years. Higher-frequency variability did take place superimposed on the long-term Pliocene SST variability that we focus on, e.g. exemplified by orbital-scale variability visible in the original raw datasets (Fig. 2). Multidecadal variability is, however, not resolved for any of the relevant Pliocene sites, and the existing age constraints are not good enough to investigate SST anomaly relations at such timescales or even at orbital scales. While not a focus for this study, we note that the amplitude of the Pliocene SST anomalies, from about 1 ∘C to close to 3 ∘C, are larger than the observation-based anomalies that are mostly less than 0.5 ∘C (Fig. 6 and 8).

A warm anomaly in the Iceland Sea corresponding with no change in the North Atlantic and the Norwegian Sea is documented for the Pliocene, between 3.43 and 3.23 Ma (Fig. 8). The idealized experiments show that buoyancy changes due to a weak atmospheric warming under constant freshwater forcing (G1 to REF-1 and G2 to REF-3) cause a warm Iceland Sea SST anomaly that differs from the North Atlantic and the Norwegian Seas, where, strictly speaking, no change is seen (Table 1 and Fig. 9a, c). Pliocene air temperatures for the Nordic Seas domain are largely unknown. No information exists from Iceland (Verhoeven et al., 2013). Over Norway, a colder air temperature is indicated between 3.43 and 3.23 Ma, mirroring the colder Norwegian Sea SSTs (Table 2) (Panitz et al., 2018). The data from Panitz et al. (2018) thus suggest a cooling over the Norwegian Sea in contrast to how the idealized experiments result in a warm anomaly in the Iceland Sea corresponding with no change in the North Atlantic and the Norwegian Sea as a response to a weak warming over the full Nordic Seas. Available data therefore cannot confirm that buoyancy change due to warmer atmospheric temperatures over the Nordic Seas may have caused the Iceland Sea SST anomaly to differ from the North Atlantic and the Norwegian Sea.

The idealized experiments that reproduce this warm Iceland Sea and cold North Atlantic and Norwegian Sea SST anomaly relation are associated with a reduction in all the following selected output parameters relative to the respective reference experiment: the density gradient between the northern and southern basin, the velocity of the inflow over the sill, the velocity of the Norwegian Sea boundary current, the heat transport over the sill, and the maximum overturning streamfunction at the sill. Existing Pliocene data indicate a weakened overturning circulation (smaller %NADW contribution) relative to the Pliocene mean between 3.43 and 3.23 Ma (Table 2) despite the Norwegian Sea being well ventilated down to intermediate depths (Risebrobakken et al., 2016) (Table 2). In line with a weakened overturning circulation, a strong Equator-to-pole North Atlantic meridional SST gradient existed due to the relatively cold North Atlantic and Norwegian Sea (Table 2; Fig. S1 in the Supplement). While the Arctic Sea ice cover was still smaller than today (Knies et al., 2014), the constant presence, rather than appearance, of IP25 in the Iceland Sea (Clotten et al., 2018) suggests more available freshwater in the form of spring sea ice relative to the intervals with spatially coherent SST anomalies (Table 2). There was, however, still much less sea ice than today. While not directly comparable, the overall reduction of the overturning at the sill in the idealized experiment seems consistent with a reduced %NADW for this period in the Pliocene (Table 2). The presence of seasonal sea ice, or sea ice transported to the Iceland Sea, suggests a slight freshening and thereby strengthened stratification in the Iceland Sea. Such a change in stratification may lead to reduced dense-water formation and may thereby weaken the NADW formation. The existence of sea ice in the Iceland Sea during the warm Pliocene suggests somewhat enhanced seasonal contrasts. Tests of different model sensitivities and freshwater-forcing scenarios using the Earth system model LOVECLIM have shown that AMOC could be more sensitive to freshwater forcing under warm interglacial climate states with large seasonal contrasts than under warm climate states with a weaker seasonal contrast (Blaschek et al., 2015). However, Blaschek et al. (2015) saw such an AMOC response only when freshwater reached the Labrador Sea convection site and not for future scenarios where less freshwater or sea ice is likely to impact the Labrador Sea. For parts of this Pliocene interval, 3.43–3.23 Ma, some seasonal sea ice existed in the Labrador Sea (Clotten, 2017). Hence, the results of Blaschek et al. (2015) show the same direction in terms of AMOC change as seen in the idealized experiments and from Pliocene data for the warm Iceland Sea and cold North Atlantic and Norwegian Sea SST anomaly relation driven by a buoyancy change due to weak atmosphere warming. A weakened ocean circulation may, again, bring less heat and salt into the Norwegian Sea and, by continuation, the Iceland Sea, further strengthening the stratification.

In the idealized experiments, a non-coherent Iceland Sea SST anomaly scenario is also obtained through strongly increased freshwater-driven buoyancy forcing under constant atmospheric temperatures (P4 to REF-2). However, the resulting response is then a cold SST anomaly in the Iceland Sea and warm anomalies in the North Atlantic and the Norwegian Sea, the opposite situation of what is seen in the Pliocene reconstructions (Figs. 6, 7, and 8).

Hence, from the idealized experiments, we manage to set up a representation of the warm Iceland Sea and cold North Atlantic and Norwegian Sea anomaly scenario. However, the link between the reconstructed background climate and oceanography and the comparable output parameters from the idealized experiments is not straightforward, or relevant data do not exist, emphasizing the need for more data for further evaluation.

A positive SST change (warming) in the Norwegian and Iceland seas corresponding with a small (close to zero) negative SST change (cooling) in the North Atlantic is seen at the end of the century (2068–2098) in CMIP6 future projections, depending on the scenario used (Fig. 7). NorESM2-MM is the least sensitive of the models, and for that model there are still spatially coherent SST anomalies at the end of the century for the SSP126 experiment, showing that the sensitivity of the model plays a role. The SST anomaly relation with the Iceland Sea and Norwegian Sea being different from the North Atlantic found in CMIP6 projections is part of a transient response to an imposed CO2 forcing. Strictly speaking, this case is therefore not representative for an equilibrium or quasi-equilibrium situation, as discussed for all the other SST anomaly relations.

A similar SST anomaly relation, with the North Atlantic being different from the Norwegian and Iceland seas, emerges in the idealized experiments for an intermediate to strong atmospheric warming over the Nordic Seas under constant freshwater (Table 1 and Fig. 9; G2–G4 to REF-1 and G2 to REF-3). Changing the buoyancy by increasing the air temperature over the Nordic Seas reflects what may happen when the atmospheric CO2 concentrations increase in the CMIP6 scenarios; Arctic amplification will entail a larger temperature change over the Nordic Seas than over the North Atlantic.

All these idealized experiments where buoyancy is changed by increasing the air temperature over the Nordic Seas are associated with a reduction in the heat transport over the sill, the maximum overturning streamfunction at the sill, the velocity of the Norwegian Sea boundary current, and the density gradient between the northern and southern basin relative to the respective reference experiments. The AMOC is also weakened compared to the historical level for all CMIP6 scenarios (Weijer et al., 2020) (Table 2). While global SSTs are overall higher for the SSP126 scenario (Table 2), the subpolar North Atlantic cools and freshens (Kwiatkowski et al., 2020). For SSP245, increased export of liquid freshwater is seen in the Fram Strait (Zanowski et al., 2021).

The same SST anomaly relations as found in the CMIP6 future scenario runs were previously identified for CMIP5 future projections (Alexander et al., 2018; Nummelin et al., 2017) and in the Grand Ensemble of the MPI-ESM1.1 climate model (Keil et al., 2020). Keil et al. (2020) show that the heat import to the North Atlantic, associated with a weakening of the low-latitude AMOC, decreases consistently, in parallel with an increased heat transport over the sill set up by a corresponding change in the high-latitude overturning and the subpolar gyre circulation, which is in line with Alexander et al. (2018) and Nummelin et al. (2017). As described above, our idealized experiments show that the SST anomaly relation, with the North Atlantic being different from the other two regions, is associated with a reduced heat transport over the sill, which appears to be opposite to the results from simulated future projections. We stress again that the result from the idealized experiments refers to equilibrium conditions, whereas the CMIP5 and 6 future projections show transient changes to enhanced CO2 forcing and thus have not reached an equilibrium. This is exemplified by the results from Keil et al. (2020), where the ocean heat transport over the sill first increases and then slightly decreases. Therefore, we find it hard to conclude what is causing this SST anomaly relation.

Independent of the exact cause, the indirect effects of both a strong increase in atmospheric CO2 concentrations (CMIP5 and 6 future projections) and increased atmospheric temperatures over the Nordic Seas (idealized experiments) may break the expectation of spatially coherent SST anomalies and set up a non-coherent SST anomaly relation between the North Atlantic (cold) and the Norwegian and Iceland seas (warm).

This SST anomaly relation is not seen during the investigated Pliocene time interval. However, we find it interesting that, despite the large differences in timescales involved (multidecadal versus hundreds of thousands of years), the strongest SST anomaly seen in the future (up to 2 ∘C between 2068 and 2098 for the Norwegian and Iceland seas in CNRM-ESM2-1; Fig. 7) is comparable to the amplitude of the Pliocene SST anomalies (Fig. 8), both of which are larger than the amplitude of the anomalies seen in the instrumental observations (less than 0.8 ∘C; Fig. 6). The amplitude of the future changes depends on both the chosen model and the scenario; these large anomalies are seen for SSP126 in CNRM-ESM2-1 and for SSP585 in NorESM. The CO2 forcing of SSP126 (445 ppm by 2100) is comparable to the high end of the Pliocene CO2 range (427 ppm) (Meinshausen et al., 2020; de la Vega et al., 2020), suggesting that the amplitude of the SST anomalies of the North Atlantic and the Nordic Seas is set by the atmospheric CO2 level and that the response time may be as short as within a century.

At two times during the Pliocene, 3.63–3.93 and 4.73–4.93 Ma, the Norwegian Sea SST anomaly differed from both the Iceland Sea and the North Atlantic (Fig. 8). Between 4.73 and 4.93 Ma, a cold SST anomaly was seen in the Norwegian Sea, contemporaneous with a warm anomaly in the Iceland Sea and the North Atlantic. The later period, 3.63 to 3.93 Ma, represents the opposite situation, with a warm Norwegian Sea anomaly corresponding to a cold SST anomaly in the Iceland Sea and the North Atlantic (Fig. 8). None of our idealized experiments resulted in an SST anomaly relation where the Norwegian Sea was different from the other two regions.

During both these Pliocene periods, the %NADW was close to the Pliocene mean strength (Table 2; Fig. S2 in the Supplement). For both cases, the intermediate-depth Norwegian Sea ventilation was stronger than in the periods with spatially coherent SST anomalies and even stronger when the Norwegian Sea was warm relative to the Pliocene mean (Table 2). The main difference between the cold (4.73 to 4.93 Ma) and warm (3.63 to 3.93 Ma) Norwegian Sea anomaly situations was that more freshwater entered the Nordic Seas from the Arctic Ocean in the warm scenario, following the opening of the Bering Strait, and that traces of sea ice were present in the Iceland Sea (Table 2). In addition, the Norwegian Sea intermediate- and upper-water-column ventilation was stronger during the warm-anomaly situation.

Since none of our idealized experiments show this SST anomaly relation, we look further into this case from a conceptual point of view. Two main oceanographic features hold the potential to change the expected spatial coherence at equilibrium between the three locations, namely a change in ocean circulation and advective pathways that weaken the advective interlinkage or a change in water column stratification in one of the regions, e.g. related to an anomalous influx of surface freshwater. A change in the advective pathway could for example be the case if the Iceland Sea is more under the influence of the Irminger Current bringing Atlantic Water to the Iceland Sea directly through the Denmark Strait rather than the water eventually reaching the Iceland Sea via the Norwegian Atlantic Current. A change in the water column stratification, e.g. in the Iceland Sea, may take place if the surface water in the Iceland Sea is more under the influence of the East Greenland Current and thus the polar domain.

Following this conceptual framework, a cold SST anomaly in the Norwegian Sea corresponding to a warm North Atlantic and Iceland Sea SST anomaly may result from a weakened Norwegian Atlantic Current compensated for by a strong Irminger Current. The dominant advective influence of the Nordic Seas is the eastern inflow via the Norwegian Atlantic Current. However, even if anomalies tend to persist throughout the Nordic Seas advective loop, the water that, at the end of this loop, travels south via the Greenland and Iceland seas will qualitatively be cold, as is the case with the water flowing out of the Nordic Seas through the Denmark Strait (e.g. Eldevik et al., 2009; Eldevik and Nilsen, 2013). The Irminger Current, on the other hand, is a warm inflow that directly influences the Iceland Sea, where it largely overturns locally to overflow through the Denmark Strait where it entered (e.g. Våge et al., 2011). A stronger Irminger Current inflow can thus be expected to leave an anomalous warm signature in the Iceland Sea – simply more warm water brought directly into the mix – independent of the anomalous state of the Norwegian Sea and the Norwegian Atlantic Current. Based on existing information, we cannot verify whether or not this was the case between 4.73 and 4.93 Ma.

In contrast, the existence of a warm anomaly in the Norwegian Sea corresponding with cold anomalies in the subpolar North Atlantic and in the Iceland Sea could, in theory, result from a strengthened or expanded East Greenland Current, increasing the fraction of cold polar water reaching the Iceland Sea (Rudels et al., 2005) and the North Atlantic (Dickson et al., 1988). Admittedly, the effect would need to be quite substantial to affect the North Atlantic proper (this is nevertheless what is implied by the common attribution of the hydrographic impact of the great salinity anomaly; Dickson et al., 1988). In general, the state of the subpolar North Atlantic tends to relate more to the larger-scale forcing or subtropical–subpolar-gyre features (e.g. Hátún et al., 2005; Reverdin, 2010), which, in this case, would be aligned to leave the region anomalously cold. The existence of sea ice in the Iceland Sea (Clotten et al., 2019), combined with the effect of enhanced freshwater supply from the Arctic as a response to the fully opened Bering Strait (De Schepper et al., 2015) (Table 2), may lend support to the occurrence of a strengthened or expanded East Greenland Current at the time of a warm Norwegian Sea SST anomaly and cold North Atlantic and Iceland Sea SST anomalies. This interval was quite similar to the cold spatial-coherence situation (Sect. 4.1; Table 2), except that the cold spatial-coherence case was associated with a higher %NADW and weaker ventilation of the upper Norwegian Sea water column. We suggest that changes in the Arctic freshwater balance, and consequently a strengthened East Greenland Current and/or stratification of the Iceland Sea, may be a likely scenario for the cold Iceland Sea and North Atlantic and warm Norwegian Sea anomaly case.

In the MITgcm set-up used, buoyancy is changed by a freshwater change that is evenly spread over the Nordic Seas, and the ocean currents in the interior basin are not very well represented (e.g. the Irminger Current). Combined, this may explain why we do not detect this SST anomaly relation through the idealized experiments. Alternatively, an even-stronger buoyancy change might be needed to set up a similar response for the idealized experiments.