Expanding Influence of Atlantic and Pacific Ocean Heat Transport on Winter Sea-Ice Variability in a Warming Arctic

Abstract

The gradual anthropogenic-driven retreat of Arctic sea ice is overlaid by large natural (internal) year-to-year variability. In winter, sea-ice loss and variability are currently most pronounced in the Barents Sea. As the loss of winter sea ice continues in a warming world, other regions will experience increased sea-ice variability. In this study, we investigate to what extent this increased winter sea-ice variability in the future is connected to ocean heat transport (OHT). We analyze and contrast the present and future link between Pacific and Atlantic OHT and the winter Arctic sea-ice cover using simulations from seven single-model large ensembles. We find strong model agreement for a poleward expanding impact of OHT through the Bering Strait and the Barents Sea under continued sea-ice retreat. Model differences on the Atlantic side can be explained by the differences in the simulated variance of the Atlantic inflows. Model differences on the Pacific side can be explained by differences in the simulated strength of Pacific Water inflows, and upper-ocean stratification and vertical mixing on the Chukchi shelf. Our work highlights the increasing importance of the Pacific and Atlantic water inflows to the Arctic Ocean and highlights which factors are important to correctly simulate in order to capture the changing impact of OHT in the warming Arctic.

Key Points

• Future climate model projections show a poleward shifted impact of Atlantic and Pacific Ocean heat transport on winter sea-ice variability

• Models with a larger variance of Atlantic inflows simulate a larger influence of Atlantic heat transport on sea ice

• Models with a stronger volume transport and downstream stratification simulate a larger influence of Pacific heat transport on sea ice

Plain Language Summary

The winter sea-ice cover in the Arctic is retreating with global warming, but with a lot of variability from year to year. Some of this variability is determined by how much oceanic heat is transported into the Arctic Ocean via the Fram Strait, Barents Sea, and Bering Strait. We explore how this link between oceanic heat transport and sea ice will change in the future when the sea ice retreats further into the Arctic Ocean. We compare several climate models and find that most of them show a northward expanding footprint of heat transport through the Barents Sea and the Bering Strait. How much these oceanic transports still affect the future sea ice depends on far the sea ice retreats, changes in the inflowing waters, and the vertical stability of the upper layer in the Arctic Ocean.

1 Introduction

The recent retreat of the Arctic sea-ice cover is overlaid by strong variability on interannual to decadal time scales, particularly during the winter months (Årthun et al., 2019; England et al., 2019). This natural (internal) variability impacts our estimates of the forced response of sea ice to global warming and is a large source of uncertainty for projections of the sea-ice cover (Bonan et al., 2021; Swart et al., 2015). In winter, a large part of the internal variability is driven by variable transport of ocean heat into the Arctic Ocean (Carmack et al., 2015; Docquier & Königk, 2021; Polyakov et al., 2020). There are three main gateways. Water from the Nordic Seas—and the Atlantic Ocean upstream—flows into the Arctic Ocean through the Fram Strait and the Barents Sea Opening (BSO, Figure 1). On the other side of the Arctic, Pacific water enters the shallow Chukchi shelf through the 50 m deep Bering Strait (Figure 1). While the water flowing into the Fram Strait typically subducts under the halocline north of Svalbard (Rudels et al., 2015) and presently has limited influence on winter sea-ice variability (Dörr et al., 2021; Lundesgaard et al., 2021), water flowing through the BSO enters the shallow Barents Sea shelf where it melts or inhibits freezing of winter sea ice in the Barents Sea and beyond (Årthun et al., 2012; Sandø et al., 2014; Schlichtholz, 2011). Oceanic heat transported through the Bering Strait over the summer season has the potential to melt large quantities of sea ice (Serreze et al., 2019; Y. Wang et al., 2021; Woodgate, 2018) and impacts the early winter sea-ice advance in the Chukchi Sea (Serreze et al., 2016).

Figure 1

Map of the Arctic Ocean. White shading and the blue line represent the mean winter sea-ice cover (50% or higher sea-ice concentration) between 1990 and 2019 (HadISST2; Titchner and Rayner (2014)). Red lines indicate the three major gateways into the Arctic Ocean on the different model grids, and the black mark the extent of the Atlantic side and Pacific side.

Over the next decades, the Arctic will likely become ice-free in summer (Notz & SIMIP community, 2020) and the sea ice in winter will retreat further into the interior Arctic Ocean, although there is substantial uncertainty about the timing and extent of the winter sea-ice loss (Årthun et al., 2021). As a consequence of the resulting changes in stratification, ocean circulation, and upper-ocean mixing, the interior upper Arctic Ocean will likely be more directly affected by changes in the Pacific and Atlantic Water inflows, an effect named Borealization (Polyakov et al., 2020), or—split up into the two regional influences—Atlantification and Pacification (Årthun et al., 2012; Dörr et al., 2021; Polyakov et al., 2017). It is therefore important to understand how far co-variability between ocean heat transport (OHT) and regional sea-ice variability will expand polewards as the sea ice retreats, not only because it will potentially affect our ability to predict sea-ice changes, but also because it is a key indicator of ongoing Borealization.

Dörr et al. (2021) used the Community Earth System Model Large Ensemble (CESM-LE) to document a future poleward expansion of the co-variability between OHT and winter sea ice. However, a key question that remains unresolved is to what extent the future connection between Atlantic and Pacific inflows and winter sea ice is set by stronger and warmer inflows, or enhanced upward fluxes of ocean heat as a result of weakened stratification (Lind et al., 2018; Polyakov et al., 2017). The sea ice is generally shielded from warm Atlantic water below by a cold layer that is strongly stratified in salinity, that is, the cold halocline (Rudels et al., 2015). Changes in the properties of this insulating layer can therefore vary the effect of warm Atlantic and Pacific waters on regional sea-ice variability. A sustained and possibly increased incursion of Atlantic waters into the Eurasian Basin throughout the century is expected (Hordoir et al., 2022; Shu et al., 2021), which could act to weaken upper-ocean stratification and increase vertical heat fluxes, and, hence, lead to winter sea ice being more strongly linked to Atlantic inflows. How inflow properties and ocean stratification impact the connection between OHT and internal sea-ice variability has not been analyzed.

Here, we compare the future connection between OHT and winter sea ice in seven single model ensembles from both the fifth (CMIP5) and the sixth (CMIP6) phase of the Coupled Model Intercomparison Project. We assess how inter-model differences in the strength and properties of the Atlantic and Pacific inflows and the representation of upper-ocean stratification, are reflected in how strongly OHT is linked to future internal sea-ice variability. This allows us to constrain the projected changes in oceanic footprint and to better understand the drivers of future Borealization of the Arctic Ocean. The analysis is structured as follows. Following an overview of the methods and model data, we compare future changes in winter sea-ice cover, inflow properties at the gateways, and Arctic Ocean stratification in Sections 3-5, respectively. We then compare changes in the link between heat transport and regional sea-ice variability and relate model differences to mean quantities in Sections 6 and 7. The discussion and summary in Section 8 conclude the study.

2 Materials and Methods

We analyze and compare monthly mean model output from seven single-model large ensembles: the CESM-LE (40 members) and GFDL-CM3-LENS (20 members) based on the CMIP5 models CESM1 and GFDL-CM3, and 5 ensembles based on the models MPI-ESM1-2-LR (10 members), MIROC6 (20 members), ACCESS-ESM1-5 (10 members), CanESM5 (10 members), and EC-Earth3 (15 members) from CMIP6 (Eyring et al., 2016). Output from GFDL-CM3-LENS and CESM-LE is available through the Multi-Model Large Ensemble Archive (Deser et al., 2020). The CMIP6 models were chosen based on a minimum member size of 10 of available output for all the relevant variables. A sufficient ensemble size is required to robustly separate internal variability from the forced signal (Milinski et al., 2020) and a threshold of 10 members represents a trade-off between robustness and the number of available models. We analyze the historical simulations and two future scenarios: a high-emissions, high warming scenario (RCP8.5 or SSP5-8.5) for all models and additionally a low warming scenario (SSP1-2.6, also referenced to as the 2°C scenario for CESM-LE) for all models except the GFDL-CM3, where this scenario was not available. Additional information about the ensemble size and future scenarios is given in table 1.

Table 1. Overview of Single-Model Ensembles Used in This Study

Model	Scenarios (ens. members)	Reference	Horiz. ocean res. north of 66°N (km)	Vertical ocean res. (upper 50 m) (m)
CESM-LE	RCP8.5 (40), 2°C (10)	Kay et al. (2015)	45	10
		Sanderson et al. (2017)	45	10
GFDL-CM3-LENS	RCP8.5 (20)	Sun et al. (2018)	55	10
MPI-ESM1-2-LR	SSP5-8.5 (10), SSP1-2.6 (10)	Mauritsen et al. (2019)	55	10.4
MIROC6	SSP5-8.5 (20), SSP1-2.6 (20)	Tatebe et al. (2019)	40	5
ACCESS-ESM1-5	SSP5-8.5 (10), SSP1-2.6 (10)	Ziehn et al. (2020)	35	10
CanESM5	SSP5-8.5 (10), SSP1-2.6 (10)	Swart et al. (2019d)	50	7.8
EC-Earth3	SSP5-8.5 (15), SSP1-2.6 (15)	Döscher et al. (2021)	50	2.8

The OHT through a section is defined as

$$$\text{OHT}=\rho {c}_{p}{\int }_{S}\mathbf{U}\left(T-{T}_{\mathit{ref}}\right)dS,$$$ (1)

where ρ = 1,025 kg m⁻³ and c_p = 4,000 J K⁻¹ kg⁻¹ are the constant density and heat capacity of seawater, U is the velocity normal to the section, T is the potential temperature and S the surface area of the section. We calculate the heat transport on the models' native grid using the model variables uo, vo, and thetao, except for CESM-LE, where we use the advective heat flux (UET and VNT). We use a reference temperature T_ref of 0°C, because the variables UET and VNT are with respect to 0°C. Choosing a different reference temperature changes the mean OHT, but it does not change the main results of this paper (not shown), that is, the connection between OHT and winter sea-ice variability, which is not strongly dependent on the reference temperature. We calculate annual mean OHT through the BSO, Bering Strait, and the Fram Strait (Figure 1). Since the flow through Fram Strait is highly bidirectional, we compute OHT only from data where water flows into the Arctic Ocean.

We use observations of sea-ice concentration from HadISST2 (Titchner & Rayner, 2014) from 1990 to 2019. Due to the shortness of the observational records of OHT, we use estimates of OHT, potential temperature, and salinity from the ocean reanalysis ORAS5 (Zuo et al., 2019) from 1990 to 2019. Upper-ocean temperatures in the Arctic from ORAS5 generally agree with observations (Li et al., 2022; Shu et al., 2021). We have furthermore compared OHT in ORAS5 with observed estimates based on mooring data in the Bering Strait (1999–2019; Woodgate & Peralta Ferriz, 2021) the BSO (1998–2016; Skagseth et al., 2020) and the Fram Strait (1998–2011; Beszczynska-Möller et al., 2012), and find that ORAS5 simulates a mean OHT similar to observations in all three gateways (Table S1 in Supporting Information S1).

We analyze monthly sea-ice concentration (model variables sic/siconc) and calculate the sea-ice area on the Pacific (Chukchi, East Siberian, Beaufort Sea, Central Arctic between 130°E and 50°W) and Atlantic side (Barents, Kara, and Laptev Sea, Central Arctic between 50°W and 130°E, Figure 1) on the native model grids by summing up the product of the grid cell area and the sea-ice concentration of all grid cells in the two regions. We compare the simulated sea-ice concentration with estimates based on satellite observations for the period 1990–2019 from HadISST2 (Titchner & Rayner, 2014).

In this study, we focus on internal sea-ice variability, which are the deviations around the forced, long-term sea-ice loss. In each of the model ensembles, all ensemble members are subject to the same external forcing, and we can robustly assess the forced signal by averaging over all members for each time step. We thus isolate internal variability from the forced signal in the model ensembles by removing the resulting ensemble mean from the raw data of each member. Since the forced signal in observations is unknown, we estimate internal variability by removing a fitted second-order polynomial from the sea ice and OHT data.

Following previous work (Årthun et al., 2012, 2019; Auclair & Tremblay, 2018; Decuypère et al., 2022; Dörr et al., 2021; Koenigk & Brodeau, 2014; Muilwijk et al., 2019; Sandø et al., 2014; Serreze et al., 2016), we compare the connection between OHT and winter sea ice by using lagged anomaly correlations. We recognize that such analysis does not allow to unequivocally infer a causal influence of OHT on sea ice. We correlate the annual mean (January-December) OHT with sea-ice concentration averaged over the following winter (November-March, starting the same year), to account for the lagged connection between OHT and sea ice (Årthun et al., 2012; Dörr et al., 2021). Sea-ice advance in the Chukchi Sea is most strongly impacted by OHT through the Bering Strait in late spring and summer (Serreze et al., 2016). The choice of annual mean OHT enables consistent averaging between the different gateways, while still accounting for the reported lags. We compare two time periods: A recent past (1990–2019) as a baseline and a future period (2050–2079) for both high and low warming scenarios. For each period, we concatenate the 30-year time series from each member (ensemble mean removed) and perform the correlations on the concatenated time series. Note that all correlations are reversed so that positive correlations indicate sea-ice loss for an increased OHT. For the analysis of transient changes in the co-variability between OHT and sea ice, we calculate the correlations for running 30-year periods from 1990 to 2019 to 2050–2079 in 1-year increments. All correlations are calculated after removing the forced signal.

To compare the stratification of the upper Arctic Ocean in the models, we follow Muilwijk et al. (2023) and calculate the annual mean ΔPE(H), which represents the energy required to fully mix the water column down to a maximum depth H. ΔPE(H) is the difference in potential energy of a water column and the same fully mixed water column down to depth H, and is defined as follows:

$$${\Delta }PE=\int \nolimits_{H}^{\mathit{surface}}g\rho (z)zdz-\int \nolimits_{H}^{\mathit{surface}}g{\rho }_{\mathit{mixed}}(z)zdz.$$$ (2)

where g is the gravitational acceleration, ρ is the water density calculated from potential temperature and salinity, and z is the depth coordinate. ΔPE is adapted from the available potential energy measure introduced by Polyakov et al. (2018). We calculate ΔPE down to 300 m, which is below the halocline in most models (Heuzé et al., 2023), and thus represents the halocline strength in the central Arctic Ocean (Muilwijk et al., 2023). If the ocean depth is lower than 300 m for a grid cell, we integrate over the entire water column. We calculate ΔPE on the original model grids, which have different vertical resolutions. As ΔPE is highly dependent on the integration depth (ΔPE ∼ H², Equation 2), we normalize it by H² to enable an inter-model comparison of the absolute stratification strength:

$$${\Delta }P{E}^{\ast }=\frac{{\Delta }PE}{{H}^{2}}.$$$ (3)

We calculate the stratification for only one member of each model since the salinity is only available for one member in GFDL-CM3. We have, however, calculated the stratification for all members in CESM-LE and find that the internal variability of 30-year average stratification is small compared to the mean, implying that using only one member provides a robust estimate of stratification changes (not shown).

3 Present and Future Winter Sea Ice

We start by evaluating the winter sea-ice cover in the different models, how it compares to observations, and how it is projected to evolve in the future. Figure 2 shows the ensemble mean winter sea-ice cover for the analyzed models, compared with that from HadISST2 for 1990–2019. On the Atlantic side, EC-Earth3 and CESM-LE simulate more ice than observed, while on the Pacific side, the GFDL-CM3 and the MIROC6 simulate less ice than observed. The MPI-ESM1-2-LR, ACCESS-ESM1-5, and the CanESM5 are broadly consistent with observations on either side, although CanESM5 has too much sea ice in the Labrador Sea.

Figure 2

Present and future winter sea-ice cover. Ensemble mean winter sea-ice cover (November-March; 50% sea-ice concentration or higher) for 1990–2019 (blue line and blue shading) and 2050–2079 (red line and pink shading) for the high emissions scenario for all models. The black dashed line shows the observed sea-ice cover from 1990 to 2019.

Forced changes in the different models are assessed by comparing the ensemble means from 2050 to 2079 with those from 1990 to 2019 (compare blue and red lines in Figure 2). For a high emissions scenario (SSP5-8.5 or RCP8.5), most models project a retreat of the winter mean ice cover toward the western Laptev Sea on the Atlantic side and the northern Chukchi Sea on the Pacific side (southern Chukchi for the MPI-ESM1-2-LR), consistent with a delayed freeze-up of the Arctic Ocean in early winter (Årthun et al., 2021; Onarheim et al., 2018). However, the CanESM5 and the GFDL-CM3 project a strong decrease in sea-ice concentration over the entire Arctic Ocean, leading to ice-free conditions during most of the winter. Under a low emissions scenario (SSP1-2.6 and 2°C), the forced changes are smaller than for the high emissions scenarios (Figure S1 in Supporting Information S1). Most models project a retreat of the mean ice edge toward the northern Barents Sea on the Atlantic and the southern Chukchi or the northern Bering Sea on the Pacific side. CanESM5 projects a much stronger sea-ice retreat than the other models on the Atlantic side.

4 Present and Future Ocean Heat Transport

Next, we assess how OHT into the Arctic from the Atlantic and Pacific Oceans is simulated in the models and how this is projected to change. The simulated evolution of OHT through the three main Arctic gateways is shown for all models in Figures 3a–3c. For the BSO, most models simulate a mean heat transport of 40–90 TW from 1990 to 2019, broadly consistent with the estimate from ORAS5 of 70 TW. The forced trend is positive for all models, but the EC-Earth3 and CanESM5 have by far the strongest trends. For future periods, the heat transport increases further in all models.

Figure 3

Present and future changes at the Arctic gateways. Time series of annual mean (a–c) heat transport, (d–f) volume transport, and (g–i) water temperature at the three gateways for ORAS5 (black line) and the seven model ensembles under a high emissions scenario. Solid lines and shading represent ensemble mean and inter-decile spread, respectively.

The OHT through the Fram Strait displays large differences across the models, values ranging from 1 TW to around 80 TW over the recent past, compared to the ORAS5 estimate of around 25 TW (Figure 3b). All models, except the MPI-ESM1-2-LR, show an increase in OHT during recent decades. Most models show a forced increase in the future under the high emissions scenario, except for GFDL-CM3 which shows a decrease.

The Bering Strait OHT ranges from 1 to 12 TW in the models, spanning the ORAS5 estimate of 6 TW (Figure 3c). All models simulate a positive forced trend in OHT over the recent past as well as for the future periods under the high emissions scenario. For all three gateways, the projected future OHT changes are similar but slightly smaller in the low emissions scenario compared to the high emissions scenario (Figure S2 in Supporting Information S1).

Changes in OHT can be due to changes in volume transport or temperature, both of which are shown in Figures 3d–3i. Note that the contributions from volume transport and mean water temperature do not constitute a full decomposition of OHT changes, as changes are also driven by joint variability in the volume transport and temperature anomalies (Shu et al., 2022). Models are broadly in agreement with observed volume transport and temperature in the BSO and to a lesser degree in the Bering Strait. In Fram Strait, the models simulate a weaker volume transport and a higher water temperature. In general, the forced increase in OHT is primarily driven by an increase in the water temperature in all models (Figures 3g–3i), especially for the Bering Strait where the volume transport decreases over time in all models (Figure 3f). We note that observations show an increase in volume transport through the Bering Strait (Woodgate, 2018), which ORAS5 does not capture. This discrepancy is common even in high-resolution models (Nguyen et al., 2020), which could be related to biases in sea surface height between the Pacific and Arctic Oceans in CMIP6 models (S. Wang et al., 2022, their Figure 8). For the BSO and the Fram Strait, some models also project strong increases in volume transport, which drive increased OHT. All cases of a forced decrease in OHT, which are most common for the Fram Strait, are driven by decreases in volume transport.

5 Present and Future Upper-Ocean Stratification

The sea-ice cover in the Arctic is generally shielded from the warmer Atlantic and Pacific waters below by a column of cold and fresher water. Next, we therefore detail future stratification changes in the Arctic since variable stratification could impact how much heat from the Atlantic and Pacific waters reaches the surface (Aagaard et al., 1981; Polyakov et al., 2018; Richards et al., 2022). We compare the mean stratification strength (ΔPE*; Equation 3) during 1990–2019 in the models and ORAS5 in Figure 4. Over the central Arctic Ocean, the stratification is stronger in the Canadian Basin than in the Eurasian Basin in ORAS5. All models struggle to capture this inter-basin difference in stratification, and, as a consequence, overestimate the stratification on the Atlantic side and underestimate it on the Pacific side. ORAS5 shows a strong stratification over most Arctic Ocean shelves, except for the southern Chukchi and Barents Sea, the inflow shelves of Pacific and Atlantic Water, respectively (white boxes in Figure 4a). There are large model differences over the shelves where some models show a stratified shelf, while others show a fully mixed water column, particularly in the southern Chukchi Sea downstream of the Bering Strait. In the Barents Sea, the inflow shelf of Atlantic Water, some models show too strong stratification, but all agree on more stratification toward the northern Barents Sea and fully mixed water columns in the southern Barents Sea.

Figure 4

Present upper-ocean stratification. Mean stratification strength (annual mean ΔPE* down to 300 m depth) between 1990 and 2019 in the ORAS5 reanalysis and the first member of each model. Note the nonlinear color bar to emphasize stratification differences on the shallow shelves. The red line denotes the 60 m isobar in the models. White boxes in (a) show the southern Chukchi Sea and the Barents Sea.

We show time series of the average stratification strength on the Pacific side and the Atlantic side until 2080 in Figure 5. We also show the southern Chukchi Sea, and the Barents Sea, as these are the inflow shelves of Pacific and Atlantic waters, respectively. All models agree that the stratification strength will increase on the Pacific side in the future. For the Atlantic side, there are large differences among the models as to whether the stratification increases, decreases, or remains the same. This is consistent with the findings of Muilwijk et al. (2023). The models agree on a constant or slightly decreasing stratification in the southern Chukchi Sea but diverge in the Barents Sea where some models show a strong increase in stratification, while others show little change.

Figure 5

Regional stratification changes. Time series of regional mean stratification strength (annual mean ΔPE* down to 300 m) in the ORAS5 reanalysis and the first member of each model, for the (a) Pacific side, (b) Atlantic side, (c) southern Chukchi Sea and (d) the Barents Sea. Faint lines denote the yearly time series, thick lines denote 30-year running means.

6 Connection Between Ocean Heat Transport and Winter Sea Ice

Having seen how the winter sea-ice cover, OHT into the Arctic, and upper-ocean stratification in the Arctic Ocean are projected to change toward the end of the century, we now assess to what extent these changes are reflected in how internal variations in OHT impact the winter sea-ice cover. Potential sources of inter-model differences in the relationship between OHT and sea-ice cover, including the role of stratification, are discussed in the next section (Section 7). Figure 6 shows anomaly correlations between the annual mean OHT and the following winter mean sea-ice concentration for the high emissions scenario. For an easier comparison of the regions where winter sea ice is linked to OHT, we only show the contours of one correlation level (r = 0.4). The chosen contour level is somewhat arbitrary and we show all contours for all models and gateways in Figures S3–S5 in Supporting Information S1. For the recent past (1990–2019), all models show a connection between the Bering Strait OHT and winter sea ice in the southern Chukchi Sea and between the BSO OHT and winter sea ice in the Barents Sea (Figure 6), consistent with ORAS5. The co-variability between the Fram Strait OHT and sea ice is limited to the northern Greenland Sea in most models, which is consistent with findings from Lundesgaard et al. (2021) showing a weak connection between Fram Strait OHT and sea-ice variability north of Svalbard.

Figure 6

The expanding co-variability between ocean heat transport (OHT) and winter sea-ice concentration. Anomaly correlation (filled contours; only 0.4 level shown) of winter mean sea ice concentration and OHT through the Fram Strait (yellow), Barents Sea Opening (blue), and Bering Strait (green) for the period 1990–2019 (dashed) and 2050–2079 (solid) under a high emissions scenario for ORAS5 and the seven model ensembles. Background shading shows the ensemble mean winter sea-ice cover (based on 50%) or higher sea-ice concentration for 1990–2019 (gray shading) and 2050–2079 (white shading).

Under a high emissions scenario, the impact of BSO and Bering Strait OHT expands toward the central Arctic Ocean in the future (2050–2079), tracking the poleward retreating sea-ice edge (Figure 2). The expanding impact is generally larger for the Bering Strait OHT on the Pacific side, where it covers parts of the Chukchi Sea, the East Siberian Sea, and the central Arctic Ocean. On the Atlantic side, the expanding impact of the BSO OHT occurs toward the Kara and Laptev Seas. Models thus roughly agree that the impact of Bering Strait and BSO OHT (i.e., Atlantification and Pacification) will converge toward the central Arctic Ocean and the North Pole. Interannual variability in winter sea ice in the Beaufort Sea and areas north of Greenland is largely unaffected by OHT in all models. This is likely due to the prevailing ice motion exporting sea ice from the central Arctic toward the Beaufort Sea by the Beaufort Gyre, thus limiting the impact of Pacific Waters on the sea-ice concentration in this region (Holland & Kimura, 2016). The future impact of the Fram Strait OHT is limited in all models. Considering the low emissions scenario, the projected changes in the relationship between OHT and the winter sea-ice cover are less pronounced than in the high emissions scenario, but, in general, show the same features (Figure S6 in Supporting Information S1).

Next, we quantify the future changes in the “footprint” of OHT, that is, the area impacted by variable OHT, by showing time series of the running 30-year mean area covered by correlations higher than 0.4 in Figures 7a–7c. We note again that the general trends in the time series are similar if we increase or decrease the correlation level to 0.3 or 0.5 (not shown). Model differences in the future footprints of OHT are larger for the Atlantic side than for the Pacific side. Models broadly agree that over time, with decreasing sea-ice cover, the footprint of the BSO OHT decreases, while the footprint of the Bering Strait OHT increases. The footprint of the Fram Strait OHT remains small. Plotting the footprint (area) of BSO and Bering Strait OHT against the mean sea-ice area on the respective sides (Figures 7d–7f) confirms that a decreasing footprint is connected to the sea-ice loss on the Atlantic side. In contrast, on the Pacific side, all models show an increasing footprint as the sea-ice cover retreats into the central Arctic Ocean, followed by a decreasing footprint for further sea-ice loss. CanESM5 and the GFDL-CM3 project the strongest decrease in the Bering Strait OHT's footprint toward 2050–2079 as a result of near ice-free winters across much of the Arctic Ocean (Figures 2b and 2e).

Figure 7

Future changes in the “footprint” of ocean heat transport (OHT). Time series of the total area north of 66° where the correlation between OHT and winter sea ice is larger than 0.4 for the (a) Barents Sea Opening (BSO), (b) Bering Strait, and (c) Fram Strait for 30-year periods between 1990–2019 and 2050–2079 for all models (plotted at the midpoints of each period, from 2005 to 2065). The same area is plotted against the winter mean sea-ice area on the Atlantic side (d, f) and the Pacific side (e). Correlation between winter mean sea-ice area on the (g) Atlantic side and OHT through the BSO, (h) Pacific side and OHT through the Bering Strait (i) Atlantic side and OHT through the Fram Strait for 30-year periods between 1990–2019 and 2050–2079 for all models. Observation-based values from ORAS5 and HadISST2 between 1990 and 2019 are marked with black stars.

We further quantify the co-variability between OHT and winter sea ice by showing time series of their running 30-year correlation (R_OHT-SIA; Figures 7g–7i). For the BSO OHT and sea-ice area on the Atlantic side, R_OHT-SIA decreases in all models (Figure 7d), consistent with the decreasing footprint over time (Figure 7a). As the sea-ice area decreases and the ice edge moves further away from the BSO, the co-variability between OHT and sea ice becomes weaker in all models. For the Bering Strait OHT and sea ice on the Pacific side, some models show a weakening connection, while others show no change over time. Here, the effect of a retreating ice edge (and thus weakening regional co-variability) is partly compensated by a strong expansion of the footprint of Pacific OHT. For Fram Strait, R_OHT-SIA is low in all models.

7 Sources of Inter-Model Differences in Oceanic Influence on Winter Sea-Ice Area

The seven single-model large ensembles analyzed here show large differences in the co-variability between OHT and winter sea ice (R_OHT-SIA; Figure 7). Next, we assess to what extent this range in R_OHT-SIA can be explained by differences in the model's mean state and variability. Specifically, we relate model differences in R_OHT-SIA between 1990 and 2079 to:

• mean sea-ice area on the Atlantic and Pacific side, and mean OHT, volume transport, and ocean temperature across the different gateways

• magnitude of internal variability (quantified as the standard deviation of anomalies over time) in sea-ice area, heat transport, volume transport, and temperature

• mean stratification (ΔPE*) in the Atlantic and Pacific region and the respective inflow shelves (Barents Sea and southern Chukchi Sea).

The relation of all factors to R_OHT-SIA is shown in Figures S7 and S8 in Supporting Information S1, whereas in Figures 8 and 9, we only show the factors that account for the largest model differences in R_OHT-SIA on the Atlantic and Pacific side, respectively. Note that we do not assess model differences for the co-variability between Fram Strait OHT and sea ice since R_OHT-SIA is low in all models (Figure 7i). For the Atlantic side, model differences in R_OHT-SIA are related to the magnitude of internal variability in sea-ice area and in the inflowing waters at the BSO (Figures 8a–8d). Models with a higher standard deviation in sea-ice area, OHT, volume transport, and temperature, have a stronger connection between BSO OHT and winter sea ice. The standard deviation of BSO temperature has the tightest relation to R_OHT-SIA. Larger variability indicates that single anomalous events are more severe, which, as a consequence, have a stronger potential impact on the sea ice downstream. Most models, but especially GFDL-CM3, ACCESS-ESM1-5, and EC-Earth3 overestimate the variability of inflows compared to ORAS5, and could thus be overestimating the link between BSO OHT and sea ice. In particular, all model ensembles overestimate the variability of OHT and volume transport compared to ORAS5. Consistent with the larger Atlantic water inflow variability, the models also overestimate sea-ice variability on the Atlantic side (Figure 8a). We find no relation between stratification on the Atlantic side, or in the Barents Sea, with R_OHT-SIA (Figure S7i and S7j in Supporting Information S1). Our results indicate that the properties and variability of the warm Atlantic Water inflow at the BSO can best explain model differences in the connection between OHT and winter sea-ice variability.

Figure 8

Sources of model differences in the co-variability between ocean heat transport (OHT) and winter sea ice on the Atlantic side. Correlation of winter sea-ice area on the Atlantic side (blue shading on map) and annual mean Barents Sea Opening (red line on map) OHT plotted against the mean (a) standard deviation in the sea-ice area on the Atlantic side, (b) standard deviation in heat transport, (c) standard deviation in volume transport, and (d) standard deviation in gateway temperature for seven different model ensembles between 1990 and 2079. Whiskers mark the total range for 30-year periods between 1990–2019 and 2050–2079. Black stars mark estimates from ORAS5 and HadISST2 between 1979 and 2019, and black whiskers mark the total range between 1979–2008 and 1990–2019.

For the Pacific side, model differences in R_OHT-SIA are related to differences in the mean and variable volume transport through the Bering Strait, as well as the magnitude of OHT variability (Figures 9a–9c). The larger the volume transport and its variability, the larger the connection between OHT and winter sea ice. This relation arises mostly from EC-Earth3, which simulates by far the largest volume transport and variability (Figure 3f), and the largest R_OHT-SIA. However, MIROC6 and ACCESS-ESM1-5 have similar volume transport and variability, but different R_OHT-SIA. This can be reconciled by looking at the stratification in the southern Chukchi Sea (the inflow shelf of Pacific Waters), which is also related to R_OHT-SIA (Figure 9d). MIROC6 has the strongest stratification and the second strongest R_OHT-SIA, while ACCESS-ESM1-5 has the weakest stratification and the second lowest R_OHT-SIA (Figures 4c and 9d). This combination of stratification and volume transport strength seems to be well captured by the bottom salinity of the Pacific Waters at Bering Strait, which we additionally analyze for the Pacific side, and which shows the strongest relation with R_OHT-SIA (Figure 9e). Here, MIROC6 and EC-Earth3 stand out as having a higher salinity than the other models, which is also more consistent with the salinity in ORAS5. We note that these two models have a significantly higher vertical resolution in the upper ocean than the other models (Table 1).

Figure 9

Sources of model differences in the co-variability between ocean heat transport (OHT) and winter sea ice on the Pacific side. As Figure 8, but for the correlation of winter sea-ice area on the Pacific side (blue shading on map) and annual mean Bering Strait (red line on map) OHT plotted against the mean (a) heat transport, (b) standard deviation in heat transport (c) standard deviation in volume transport, (d) stratification strength over the southern Chukchi Sea and (e) bottom salinity at the Bering Strait.

These results are somewhat counterintuitive because a higher bottom salinity and stronger stratification on the inflow shelf would suggest less mixing of Pacific Waters with the surface waters at the sea-ice edge, and therefore a weaker impact of OHT on sea-ice variability. One hypothesis is that if the water column on the Chukchi shelf is strongly stratified, warm Pacific Water occupying the shelf below the surface layer during summer and autumn can preserve its properties better during this period, before being mixed up in late autumn, delaying the cooling of surface waters and subsequent sea ice formation (Y. Wang et al., 2021). We test this hypothesis by comparing in Figure 10 the mean temperature and salinity during late autumn (November and December) in a section across the Chukchi shelf in MIROC6 (strongly stratified), ACCESS-ESM1-5 (weakly stratified) and ORAS5. In both ORAS5 and MIROC6, there is warm and saline Pacific Water present in the bottom layers (Figures 10a–10c and 10g–10i), which, in line with the hypothesis presented above, has the potential to delay surface cooling when the water column is mixed during the freezing season. This salinity and temperature maximum is not present in ACCESS-ESM1-5 and other models with weak stratification (Figures 10d–10f; Figure S9 in Supporting Information S1). These results can thus help to explain why the connection of Bering Strait OHT is weaker for models with less shelf stratification. An exception is EC-Earth3, which shows little stratification, but still has saline and warm Pacific Water over the Chukchi shelf (Figures S9p–S9r in Supporting Information S1) due to a large amount of Pacific Water entering the shelf (large volume transport in Figure 9a).

Figure 10

Chukchi shelf temperature and salinity. Vertical sections of average November-December temperature (a, d, g) and salinity (b, e, h) across the sections shown in (c, f, i) during 1990–2019 for MIROC6, ACCESS-ESM1-5, and ORAS5. Horizontal dashed lines indicate the vertical resolution.

8 Discussion and Conclusions

Variable OHT is at present an important source of internal winter sea-ice variability in the Arctic (Årthun et al., 2012; Dörr et al., 2021; Sandø et al., 2014; Serreze et al., 2016), but it remains uncertain how this impact will change in the future. In this study, we analyzed projected changes in the co-variability between OHT and winter sea ice using seven single-model large ensembles from CMIP5 and CMIP6. Based on these model projections, we find that the influence of Atlantic and Pacific heat transport will expand polewards into the Arctic Ocean in the future (Figure 6). The expanding impacts divide the Arctic Ocean into two regimes, with Atlantic heat transport mainly affecting sea ice in the Eurasian Arctic (west of the Laptev Sea) and Pacific heat transport affecting sea ice in the Amerasian Arctic. This regional difference in the “footprints” of Atlantic and Pacific heat transport is in agreement with Dörr et al. (2021), Richards et al. (2022), and Muilwijk et al. (2023). The connection between the Fram Strait heat transport and winter sea-ice variability is limited in all models. This suggests that the Atlantic Water subducting under the halocline west of Svalbard does not affect the sea ice on interannual timescales. On one hand, this finding is consistent with Lundesgaard et al. (2021) who found that observed sea-ice variability north of Svalbard is not driven by variability in the Atlantic inflow, but rather by local winds. On the other hand, Ivanov et al. (2016) suggest an impact of Atlantic Water temperature on sea ice northeast of Svalbard, and the models examined here might underestimate the future impact of the Fram Strait OHT because they simulate a too strong stratification and a too deep Atlantic Water core in the Eurasian Basin (Figure 4; Heuzé et al. (2023)).

In all models, future changes in the footprint of OHT can be traced back to changes in the mean sea ice state. Especially on the Atlantic side, the impact of OHT moves northwards as the ice edge moves northwards in all models (Figure 6). The area impacted by Atlantic OHT nevertheless becomes smaller (Figure 7a). On the Pacific side, the generally weakening connection between OHT and sea ice (Figure 7h) is accompanied by a strong expansion of area impacted by Pacific heat transport (Figure 7b). The area impacted by Pacific OHT only starts to decrease for a total loss of winter sea ice, something which two models indicate under a strong emissions scenario.

While future sea-ice loss explains the reduced co-variability between OHT and winter sea ice in all models, it does not explain the intermodel differences. Instead, on the Atlantic side, model differences in the connection between OHT and sea ice can best be explained by differences in the variability of the Atlantic Water inflow at the BSO (Figure 8). In general, models overestimate Atlantic Water variability compared with an ocean reanalysis product (ORAS5) and therefore might overestimate its link to sea-ice variability. Differences in the simulated stratification on the Atlantic side are not related to the direct co-variability between BSO OHT and winter sea ice (Figure S7 in Supporting Information S1). On the Pacific side, model differences are best explained by differences in the volume transport and inflow shelf stratification (Figure 9). Most models likely underestimate the present link between Bering Strait heat transport and winter sea ice due to a combination of underestimated volume transport (Figures 3f and 9a) and an underestimated stratification over the Chukchi shelf (Figure 10). The latter leads to too strong mixing of Pacific Waters with the surface polar waters, which reduces the impact of Pacific Water on the Chukchi shelf. Our results suggest that model biases in upper-ocean stratification and vertical mixing need to be reduced to accurately capture the Pacific Water flow and its interaction with sea ice on the Chukchi shelf. These biases could be reduced by increasing the vertical resolution of the upper ocean to properly resolve the shallow summer mixed layer (Rosenblum et al., 2021).

We focused in this study on projected changes under a strong emissions scenario (SSP5-8.5/RCP8.5), but our main conclusions also hold for the low emissions scenario (SSP1-2.6). The future changes in sea ice are smaller in the low emissions scenario (Figure S1 in Supporting Information S1) and the models show a smaller poleward expansion of the impact of Atlantic and Pacific heat transport (Figure S6 in Supporting Information S1). We find, however, the same sources of model differences as those identified for the high emissions scenario (Figures 8 and 9), suggesting that our results are independent of the exact strength of future warming.

The connection between OHT and sea ice is likely also affected by other factors than those investigated here. The heat transport through the BSO is for example, influenced by atmospheric variability over the Nordic Seas (Q. Wang et al., 2019; Madonna & Sandø, 2022). However, we find no relationship between the strength in atmospheric forcing (quantified as the strength of the associated sea level pressure anomaly over Svalbard) and the link between OHT and sea ice in the models (not shown). For the Pacific side, differences in the strength of the connection of sea ice to the Bering Strait heat transport could also be related to differences in the simulated pathways of Pacific Water from the Bering Strait toward the central Arctic Ocean. For example, CESM-LE struggles to accurately simulate those pathways (Lavoie et al., 2022), and the same is possibly true for other models.

We focused our analysis on the interannual variability of winter sea ice and OHT. However, OHT also affects internal sea-ice variability on longer timescales (Årthun et al., 2019). Internally driven 30-year trends in winter sea-ice area on the Atlantic and Pacific sides are significantly correlated to trends in OHT through the BSO and Bering Strait, respectively, for all models, both now and in the future (Figure 11), although much weaker in the future in some models. For externally driven trends (comparing ensemble mean trends), there is a strong connection for the Atlantic side, weakening in the future, but a much weaker connection for the Pacific side. This suggests that the long-term (externally forced) increase in oceanic heat input is a major driver for the sea-ice loss on the Atlantic side, but not the main driver on the Pacific side. However, since all models project a decrease in volume transport through Bering Strait, while observations show an increase (Woodgate, 2018), they might underestimate the future connection between long-term sea-ice loss and OHT on the Pacific side.

Figure 11

Relationship between ocean heat transport (OHT) and sea-ice trends. Scatter plot of 30-year trends in winter sea-ice area for (a, c) the Atlantic side and (b, d) the Pacific side against 30-year trends in the OHT through (a, c) the Barents Sea Opening and (b, d) the Bering Strait for (a, b) 1990–2019 and (c, d) 2059–2079. Light dots represent single members, large markers indicate each model's ensemble mean. Correlations figures are over the models' averages (R_model) and the mean correlation over each model's ensemble members (R_internal), with a range of all models given in brackets.

To identify sources of model uncertainty in future sea ice projections, several studies have sought after emergent constraints, which are simple relationships between sea-ice loss and mean quantities that are valid in a large range of models and observations (Horvat, 2021; Mahlstein & Knutti, 2012; Massonnet et al., 2012, 2018). Instead of constraining sea ice projections, where OHT is often used as a constraining variable, here we tried to constrain the future role of OHT itself. Consistent with the findings of Docquier et al. (2020), we find that the identified relationships are independent of horizontal model resolution (Table 1), although we note that the resolution might be too low to accurately capture the Fram Strait OHT in all models. The vertical resolution may however play a role as discussed above. Besides identifying the constraining factors, a result of our study is that the factors on the Pacific side are different from the ones on the Atlantic side. The expanding impact of Atlantic and Pacific heat on winter sea ice can be seen as tracers of the Atlantification and Pacification of the upper Arctic Ocean, which has important consequences for the Arctic ecosystem (Ingvaldsen et al., 2021; Polyakov et al., 2020).

Results based on future projections from coarse-resolution climate models are inherently uncertain. Accurately simulating the circulation of the Arctic Ocean and its interaction with sea ice is challenging, and CMIP5 and CMIP6 models show substantial biases in their representation of the Arctic Ocean (Heuzé et al., 2023; Khosravi et al., 2022; Muilwijk et al., 2023; Shu et al., 2019; S. Wang et al., 2022; Zanowski et al., 2021). Our study reveals large model differences in the representation of oceanic influence on sea-ice variability and highlights the processes that have to be accurately represented in current climate models to capture the impact of OHT on the future Arctic winter sea-ice cover.

Acknowledgments

All authors were funded by the Research Council of Norway projects Nansen Legacy (Grant 276730) and the Trond Mohn Foundation (Grant BFS2018TMT01). MÅ is also supported by the Research Council of Norway project Overturning circulation in the new Arctic (Grant 335255). ABS is supported by the Research Council of Norway project CASINO (Grant 325665). We acknowledge the World Climate Research Programme, which, through its Working Group on Coupled Modelling, coordinated and promoted CMIP6. We thank the climate modeling groups for producing and making available their model output, the Earth System Grid Federation (ESGF) for archiving the data and providing access, and the multiple funding agencies who support CMIP6 and ESGF. We thank the US CLIVAR Working Group on Large Ensembles for providing large ensemble output via the Multi-Model Large Ensemble Archive. Furthermore, we thank the CESM Large Ensemble Community Project for making their data publicly accessible. We thank four anonymous reviewers for valuable feedback that helped improve this manuscript.

Open Research

Data Availability Statement

All data in this study are publicly available. Output from ORAS5 is available through the Copernicus Climate Change Service's Climate Data Store (Copernicus Climate Change Service, 2021). Output from the CMIP6 models MPI-ESM1-2-LR (Wieners et al., 2019a, 2019b, 2019c), MIROC6 (Shiogama et al., 2019a, 2019b; Tatebe & Watanabe, 2018), CanESM5 (Swart et al., 2019a, 2019b, 2019c), EC-Earth3 (EC-Earth Consortium (EC-Earth), 2019a, 2019b, 2019c), and ACCESS-ESM1-5 (Ziehn et al., 2019a, 2019b, 2019c) are available via the Earth System Grid Federation's CMIP6 archive (https://esgf-index1.ceda.ac.uk/search/cmip6-ceda/). Output from CESM-LE's high warming and low warming runs is available via the Earth System Grid (Climate Data Gateway, 2021; Sanderson, 2017). Output from the GFDL-CM3-LENS is available through the Multi-Model Large Ensemble Archive (Climate Data Gateway, 2019). Observed sea-ice concentration from HadISST.2.0.0 is available through the UK Met Office website (Titchner, 2020; Titchner & Rayner, 2014).