Arctic sea ice
Figures 1 and 2 show monthly mean Arctic sea-ice area and volume, respectively, averaged over 1979–2014. This period is chosen to be comparable to observations, but key results are independent of the chosen period. For all of the models used in this study, we find a year-round decrease of Arctic sea-ice area and volume with finer ocean resolution (comparing HadGEM3-LL with HadGEM3-MM/HM, ECMWF-LR with ECMWF-MR/HR, and AWI-LR with AWI-HR). This decrease is especially pronounced for ECMWF-IFS, i.e. \(-\,23\%\) to \(-\,30\%\) in area (Table 2) and \(-\,36\) to \(-\,49\%\) in volume (Table 3) over the whole period. The change in sea-ice area and volume is less clear with changing atmosphere resolution. For HadGEM3, increasing the atmosphere resolution from 100 (HadGEM3-MM) to 50 km (HadGEM3-HM) leads to lower sea-ice area and volume (Figs. 1, 2). However, sea-ice area and volume increase with finer atmosphere resolution for ECMWF-IFS (from 50 km in ECMWF-MR to 25 km in ECMWF-HR) and CMCC-CM2 (from 100 km in CMCC-HR4 to 25 km in CMCC-VHR4). For MPI-ESM, sea-ice area generally increases with higher atmosphere resolution, although the increase is relatively small and not significant over the whole time period (Fig. 1, Table 2), while sea-ice volume decreases (Fig. 2, Table 3). Thus, the implication from this sample of models is that a finer ocean resolution leads to reduced Arctic sea-ice area and volume, while the impact of atmosphere resolution is less clear. This point will be further discussed in Sect. 4.1.
Table 2 Mean differences in Arctic sea-ice area (\(10^6~\hbox {km}^2\)) between the different configurations of each model averaged over all months of the period 1979–2014, over winter months (January–March 1979–2014), and over summer months (July to September 1979–2014) Table 3 Mean differences in Arctic sea-ice volume (\(10^3~\hbox {km}^3\)) between the different configurations of each model averaged over all months of the period 1979–2014, over winter months (January to March 1979–2014), and over summer months (July to September 1979–2014) Both HadGEM3 and ECMWF-IFS models generally overestimate sea-ice area and volume during all months compared to OSI SAF (Fig. 1) and PIOMAS (Fig. 2), respectively, with an unrealistically high volume for the ECMWF-LR configuration. For HadGEM3 and ECMWF-IFS, using a finer ocean resolution provides results in better agreement with both sea-ice area from OSI SAF (Fig. 1) and sea-ice volume from PIOMAS (Fig. 2). ECMWF-MR is very close to the observed sea-ice area, and HadGEM3-HM is in good agreement with the sea-ice volume from PIOMAS. The situation is not as clear-cut for AWI-CM, CMCC-CM2 and MPI-ESM models. Both AWI-CM configurations overestimate sea-ice area in winter and underestimate it in summer compared to observations. The finer resolution (AWI-HR) is closer to OSI SAF sea-ice area during December-May, and farther away the rest of the year compared to AWI-LR (Fig. 1). CMCC-HR4 underestimates sea-ice area, while CMCC-VHR4 stays within the bounds of interannual variability of OSI SAF (Fig. 1). MPI-ESM sea-ice area agrees with OSI SAF within the bounds of interannual variability in winter and underestimates this quantity in summer. The finer resolution of MPI-ESM (MPI-XR) is closer to OSI SAF in terms of sea-ice area from November to March and June, and farther away the rest of the year compared to MPI-HR (Fig. 1). Both AWI-CM and MPI-ESM slightly underestimate sea-ice volume compared to PIOMAS, with the coarser resolutions of these two models being closer to PIOMAS during the whole year (Fig. 2). CMCC-HR4 slightly overestimates the sea-ice volume from reanalysis, while CMCC-VHR4 clearly has too high ice volume (Fig. 2).
Despite these model biases, all five models can reproduce the general behavior of the mean seasonal cycles of Arctic sea-ice area and volume compared to OSI SAF observations (Fig. 1) and PIOMAS reanalysis (Fig. 2), respectively, with a maximum in March for area and April for volume, and a minimum in August-September for area and September for volume. All HadGEM3, AWI-CM and MPI-ESM configurations, as well as the low resolution of ECMWF-IFS, overestimate the amplitude of the mean seasonal cycles of Arctic sea-ice area and volume compared to OSI SAF observations (Fig. 1) and PIOMAS reanalysis (Fig. 2), respectively. The two higher resolutions of ECMWF-IFS and the two CMCC-CM2 configurations underestimate these cycles. Using a finer resolution has different implications on the amplitude of the seasonal cycles of sea-ice area and volume for the different models. For HadGEM3, the amplitude decreases and is in better agreement with observations/reanalysis at finer ocean resolution. For ECMWF-IFS, the amplitude also decreases with finer ocean resolution, but the coarser resolution (ECMWF-LR) has an amplitude of sea-ice area that is closer to observations, while the amplitude of sea-ice volume is closer to PIOMAS with the finer resolutions (ECMWF-MR/HR). For AWI-CM, the amplitude stays relatively similar between both ocean resolutions. For CMCC-CM2, the amplitude slightly increases and is in better agreement with observations with finer atmosphere resolution. For MPI-ESM, the amplitude increases with finer atmosphere resolution, and the coarser resolution has an amplitude closer to observations.
In agreement with OSI SAF observations and PIOMAS reanalysis, the trends in Arctic sea-ice area and volume over 1979–2014 are significantly negative (\(5\%\) level) for all models and all months (Figs. 3 and 4). Compared to observations and reanalysis, the trends in sea-ice area and volume of HadGEM3-LL, ECMWF-LR and MPI-HR are generally more negative than the observed and reanalysis trends. On the contrary, the trends in area and volume of HadGEM3-HM, ECMWF-MR, ECMWF-HR, AWI-LR and MPI-XR are generally less negative than the observed and reanalysis trends. The two CMCC-CM2 configurations have trends in sea-ice volume that are more negative than PIOMAS during the whole year, while they have trends in sea-ice area that are less negative than OSI SAF in summer. AWI-HR has a trend in sea-ice volume that is less negative than PIOMAS. For all models, the trends in sea-ice area and volume are less negative with finer ocean resolution, with the exception of AWI-CM for which the trend in sea-ice area becomes more negative with finer resolution. As for mean values, the impact of atmosphere resolution on trends in area and volume is not clear, with less negative trends for HadGEM3 (comparing HadGEM3-MM and HadGEM3-HM) and MPI-ESM, and more negative trends for ECMWF-IFS (comparing ECMWF-MR and ECMWF-HR) and CMCC-CM2 (Figs. 3 and 4). The higher resolution configurations do not necessarily have sea-ice area and volume trends in closer agreement with observations and reanalysis. Comparing the mean sea-ice volume (Fig. 1) and the trend in sea-ice volume (Fig. 3), we find that models with lower mean sea-ice volume generally have less negative trends in sea-ice volume. This can be explained by the ice growth-thickness feedback, i.e. models with thinner sea ice have larger ice-growth rates, partly limiting sea-ice melting (Bitz and Roe 2004). Thus, models with thinner sea ice, such as the higher ocean resolution versions of the models used here, have a slower loss of ice volume, which would explain the reduced negative trends at finer ocean resolution.
Mean March sea-ice thickness decreases with finer ocean resolution in all regions of the Arctic for ECMWF-IFS and AWI-CM, while it stays relatively similar for HadGEM3 (Fig. 5). With finer atmosphere resolution, the mean March thickness decreases for MPI-ESM and increases for CMCC-CM2 (Fig. 5). These results are valid for all months of the year, as summarized in Table 4. Compared to ICESat observations (Fig. 5l) and PIOMAS reanalysis (Fig. 5i), HadGEM3, ECMWF-IFS and CMCC-CM2 configurations overestimate the mean sea-ice thickness, while AWI-CM and MPI-ESM underestimate it. The highly overestimated thickness from ECMWF-LR (Fig. 5b) combined with too high sea-ice area (Fig. 1) lead to the unrealistically high sea-ice volume of this model configuration (Fig. 2). The biases in Arctic sea ice simulated by ECMWF-LR are partly explained by excessive ice growth due to negative biases in longwave and shortwave cloud radiative forcings over the Arctic (Roberts et al. 2018).
Table 4 Mean differences in Arctic sea-ice thickness (m), averaged over the area north of \(70^\circ \hbox {N}\), between the different configurations of each model averaged over all months of the period 1982–2014, over winter months (January to March 1982–2014), and over summer months (July to September 1982–2014) The location of the Arctic sea-ice edge (defined as the isoline where sea-ice concentration is \(15\%\)) is generally better represented at finer resolution with ECMWF-IFS, HadGEM3 and CMCC-VHR4 compared to OSI SAF observations (Fig. 5). In the low-resolution configurations of ECMWF-IFS (Fig. 5b) and HadGEM3 (Fig. 5a), the sea-ice edge typically extends too far south in both Bering and Labrador Seas. The situation is more nuanced for AWI-CM and MPI-ESM, with an improvement at finer resolution in some cases (e.g. March sea-ice edge in Bering Sea in AWI-HR, Fig. 5f) and a worsening in other cases (e.g. September sea-ice edge in AWI-HR, Fig. 5f).
All the results of this section are based on historical runs (hist-1950). When we use control runs (control-1950), HadGEM3, ECMWF-IFS and AWI-CM also exhibit lower sea-ice area and volume with finer ocean resolution, while MPI-ESM has higher sea-ice area with higher resolution, in a similar way as for hist-1950 runs. Thus, control-1950 runs confirm our results based on hist-1950 runs. The fact that control-1950 and hist-1950 runs provide similar results means that our findings are independent of the presence of time-evolving external climate forcings.
In the models studied, the main impacts of model resolution on Arctic sea ice (area, volume, thickness, edge) are summarized below:
-
sea-ice area and volume decrease with finer ocean resolution, while the impact of atmosphere resolution is less clear (Figs. 1, 2, Tables 2, 3);
-
a finer ocean resolution leads to improved sea-ice area and volume compared to observations and reanalysis for HadGEM3 and ECMWF-IFS (Figs. 1, 2);
-
a finer resolution leads to worsened sea-ice volume compared to reanalysis for AWI-CM, CMCC-CM2 and MPI-ESM (Fig. 2);
-
a finer ocean resolution leads to a decrease in the amplitude of mean seasonal cycles of sea-ice area and volume for HadGEM3 and ECMWF-IFS and no change for AWI-CM (Figs. 1 and 2);
-
the trends in sea-ice area and volume are less negative with finer ocean resolution (except the trend in sea-ice area of AWI-CM) (Figs. 3, 4);
-
the mean sea-ice thickness clearly decreases with finer ocean resolution for ECMWF-IFS and AWI-CM, while it stays relatively similar for HadGEM3 (Fig. 5, Table 4);
-
the location of the sea-ice edge is better represented at finer resolution with ECMWF-IFS, HadGEM3 and CMCC-CM2 (Fig. 5);
-
the results from control runs (control-1950) are comparable to historical runs (hist-1950).
North Atlantic OHT
Figure 6 presents the mean northward OHT in the Atlantic averaged over 1950–2014. For all model configurations, the latitudinal variation of OHT follows the observed profiles (hydrographic measurements and estimates), but the OHT is generally underestimated by models between \(20^\circ \hbox {S}\) and \(30^\circ \hbox {N}\) and overestimated at higher latitudes. OHT model underestimation at low latitudes and overestimation at high latitudes reflects insufficient heat loss to the atmosphere between these two latitudes. The insufficient North Atlantic heat loss in models is an important topic of research, which may be partially addressed as resolution increases to eddy-resolving scale (Roberts et al. 2016).
Enhanced ocean resolution implies increased poleward OHT at all latitudes for HadGEM3 and ECMWF-IFS (Fig. 6), in closer agreement with the OHT estimates from Trenberth and Fasullo (2017). The two AWI-CM configurations follow each other closely, although the OHT is globally higher at finer resolution (especially around \(24^\circ \hbox {N}\) and \(45^\circ \hbox {N}\)). A finer atmosphere resolution leads to higher Altantic OHT for HadGEM3 and CMCC-CM2 (although CMCC-VHR4 has lower OHT at high latitudes compared to CMCC-HR4), lower OHT for MPI-ESM, and almost no change for ECMWF-IFS (Fig. 6). Note that recent studies show that HadGEM3-GC2 (Hewitt et al. 2016) and ECMWF-IFS (Roberts et al. 2018) present smaller differences in OHT when only the atmosphere resolution is varied compared to ocean resolution.
Most model configurations underestimate the mean OHT observational estimate of \(1.21 \pm 0.34\) petawatts (PW) from the RAPID-MOCHA array (\(26.5^\circ \hbox {N}\) in the Atlantic Ocean), averaged over 2005-2014 (Table 5). A finer ocean resolution brings the models in better agreement with these observations (from HadGEM3-LL to HadGEM3-MM/HM; from ECMWF-LR to ECMWF-MR/HR; from AWI-LR to AWI-HR). A finer atmosphere resolution has different implications, with higher OHT at \(26.5^\circ \hbox {N}\) for HadGEM3 (from HadGEM3-MM to HadGEM3-HM), ECMWF-IFS (from ECMWF-LR to ECMWF-MR) and CMCC-CM2, and lower OHT at \(26.5^\circ \hbox {N}\) for MPI-ESM. Compared to the mean observed OHT at the BSO of \(49 \pm 21\) terawatts (TW; averaged over 1998–2014), HadGEM3 and ECMWF-IFS clearly underestimate the observed value at low ocean resolution (\(1^\circ\)) and are in relatively good agreement with observations at finer ocean resolution (\(0.25^\circ\)) (Table 5). Increasing the atmosphere resolution does not lead to an improvement of OHT at the BSO compared to observations for HadGEM3 and ECMWF-IFS, while it does for CMCC-CM2 and MPI-ESM.
Table 5 Mean observed and modeled OHT at the RAPID-MOCHA array (\(26.5^\circ \hbox {N}\) in the Atlantic Ocean) and the Barents Sea Opening (BSO; \(20^\circ \hbox {E}\), 71.5–\(73.5^\circ \hbox {N}\)) averaged over 2005–2014 and 1998–2014, respectively Overall, trends in Atlantic OHT at \(50^\circ \hbox {N}\), \(60^\circ \hbox {N}\) and \(70^\circ \hbox {N}\) from 1979 to 2014 decrease with finer ocean resolution (Table 6). The situation is more nuanced with atmosphere resolution, with a decreasing trend with resolution for HadGEM3 and MPI-ESM and an increasing trend for ECMWF-IFS and CMCC-CM2. However, the trend in OHT at \(50^\circ \hbox {N}\) is significant (\(5\%\) level) only for 6 model configurations (out of 12) and the OHT trend at \(60^\circ \hbox {N}\) is significant only for 2 model configurations (Table 6). Furthermore, note that trends at \(50^\circ \hbox {N}\) are mostly negative. On the contrary, most model configurations show a significant positive trend in OHT at \(70^\circ \hbox {N}\) (Table 6). For HadGEM3-MM and HadGEM3-HM, the OHT trend at \(70^\circ \hbox {N}\) is not significant but positive. These positive OHT trends at \(70^\circ \hbox {N}\) coincide with the negative trends in Arctic sea-ice area (Fig. 3) and volume (Fig. 4) over the same time period (1979–2014), highlighting the link between OHT at sufficiently high latitude (\(70^\circ \hbox {N}\) in our case) on the one hand and Arctic sea-ice area and volume on the other hand. The only model configuration for which the trend in OHT at \(70^\circ \hbox {N}\) is negative is ECMWF-MR, but it is not significant (Table 6). Note that this specific model configuration presents the least negative trends in sea-ice area (Fig. 3) and volume (Fig. 4). In summary, a finer ocean resolution generally results in less positive trends in OHT at \(70^\circ \hbox {N}\) (Table 6) and less negative trends in Arctic sea-ice area (Fig. 3) and volume (Fig. 4).
Table 6 Trends in OHT at \(50^\circ \hbox {N}\), \(60^\circ \hbox {N}\) and \(70^\circ \hbox {N}\) (TW \(\hbox {decade}^{-1}\)) computed over all years of the period 1979–2014 Spatial analysis of SST, ocean surface velocity and sea-ice edge provides insight into the potential links between OHT and Arctic sea ice. More specifically, the mean SST in the North Atlantic Ocean (especially between 40 and \(70^\circ \hbox {N}\)) increases with finer ocean resolution, as illustrated by HadGEM3 (Fig. 7a, d, j), ECMWF-IFS (Fig. 7b, e, k) and AWI-CM (Fig. 7c, f, l). The role of atmosphere resolution is again more complex: for example, for MPI-ESM, the mean SST decreases with finer atmosphere resolution (Fig. 7g, h, m). The impact of resolution on sea-surface velocity is even more explicit, with an overall intensification and better position of the North Atlantic currents with finer ocean resolution for HadGEM3, ECMWF-IFS and AWI-CM, while the velocity decreases with MPI-ESM (Fig. 8). In Figs. 7 and 8, we only show the March SST and surface velocity fields (averaged over 1982–2014), respectively, but these statements are valid for all months of the year. The higher SST and ocean surface velocity of HadGEM3-MM and ECMWF-HR (compared to HadGEM3-LL and ECMWF-LR, respectively) lead to a retreated sea-ice edge in the North Atlantic Ocean (Figs. 7d, e and 8d, e, compared to Figs. 7a, b and 8a, b, respectively). On the contrary, the lower SST and surface velocity of MPI-XR compared to MPI-HR results in a sea-ice edge located farther south (Figs. 7h, 8h, compared to Figs. 7g and 8g, respectively). Compared to observed SST and ocean surface velocity, the enhanced ocean resolution clearly improves the model results (Figs. 7, 8, Table 7). Especially, we note that an eddy-permitting ocean resolution (\(\sim 0.25^\circ\)) provides SST and ocean surface velocity in better agreement with observations compared to a coarser resolution, with a clear model underestimation of these two fields with an ocean resolution of \(\sim 1^\circ\). Note that further improvements are expected at higher ocean resolution (i.e. \(1/12^\circ\) or higher), which would allow to resolve ocean Rossby radius at mid and high latitudes and coastal regions.
Table 7 Mean March SST and ocean surface velocity in the central North Atlantic Ocean (domain: 20–40\(^\circ \hbox {N}\), 60–\(20^\circ \hbox {W}\)) averaged over 1982–2014, corresponding to the models and observations used in Figs. 7 and 8, respectively The complex ocean surface circulation of specific regions, such as the Barents Sea, is better represented with refined model ocean resolution down to \(0.25^\circ\) or with a variable-resolution mesh (Fig. 9). As the ocean resolution increases, a detailed path of the surface circulation emerges in Barents Sea, and the currents intensify. This is especially clear for the HadGEM3 and ECMWF-IFS models. Surface velocity field at \(1^\circ\) ocean resolution is very weak, reaching only few cm \(\hbox {s}^{-1}\) in HadGEM3-LL (Fig. 9a) and ECMWF-LR (Fig. 9b). At \(0.25^\circ\) ocean resolution (HadGEM3-MM, ECMWF-HR; Fig. 9d,e), the surface current paths are clearly identified. In particular, the Norwegian Atlantic Current, flowing poleward, splits into two distinct branches: one part flows eastward through the BSO and the other continues toward Fram Strait (as West Spitsbergen Current). Within the Barents Sea, we clearly distinguish the main counterclockwise circulation and the current system between Frans Josef Land and Novaya Zemlya. The position of the sea-ice edge retreats northward as the model resolution is refined for ECMWF-IFS (Fig. 9b,e), in line with a stronger surface current flowing through the BSO. For AWI-CM, the surface velocity increases with finer resolution (Fig. 9c,f), despite the fact that AWI-LR and AWI-HR have approximately the same ocean resolution in the Barents Sea (\(\sim 25\) km). A detailed study of the AWI-CM model shows that AWI-LR overestimates the convection in Greenland–Iceland–Norwegian (GIN) Seas, which in turn reduces the Atlantic water inflow into the Barents Sea (Sein et al. 2018). For MPI-ESM, the ice edge extends farther south with finer resolution as the surface velocity decreases (Fig. 9g,h).
The spatial details of sea-ice concentration in the Barents Sea are better captured with a sufficiently high resolution for HadGEM3 and ECMWF-IFS (Fig. 10). HadGEM3-LL has a very sharp gradient in sea-ice concentration from the western to eastern Barents Sea (Fig. 10a) compared to observations (Fig. 10i), while HadGEM3-MM has a smoother and more realistic pattern (Fig. 10d). ECMWF-LR clearly overestimates sea-ice concentration in the whole Barents Sea (Fig. 10b), while ECMWF-HR is much closer to observations (Fig. 10e). For AWI-CM, the mean sea-ice concentration in the Barents Sea computed from AWI-LR is closer to observations compared to AWI-HR, which has slightly more ice than AWI-LR (Fig. 10c, f). However, the ocean resolution in this region is relatively similar between both configurations; therefore, differences might be due to other factors (e.g. stronger ocean currents in AWI-HR compared to AWI-LR, Fig. 9c, f). The two MPI-ESM configurations do not differ substantially in terms of sea-ice concentration in the Barents Sea, despite some differences west of Svalbard and off Novaya Zemlya coast (Fig. 10g, h).
As for Arctic sea ice (Sect. 3.1), the results of this section are based on historical runs (hist-1950). The use of control runs (control-1950) leads to similar results, i.e. a finer ocean resolution results in higher poleward OHT from the North Atlantic Ocean (not shown).
In the models studied, the main effects of model resolution on North Atlantic OHT and related fields (SST and ocean surface velocity) are summarized below:
-
enhanced ocean resolution implies increased poleward OHT, while the role of atmosphere resolution is less clear (Fig. 6, Table 5);
-
trends in Atlantic OHT at \(50^\circ \hbox {N}\), \(60^\circ \hbox {N}\) and \(70^\circ \hbox {N}\) decrease with finer resolution, with a significant positive trend in OHT at \(70^\circ \hbox {N}\) for most model configurations (Table 6);
-
the mean SST and ocean surface velocity in the North Atlantic Ocean increase with finer ocean resolution (Figs. 7 and 8, Table 7);
-
the complex ocean surface circulation of the Barents Sea requires an ocean resolution of at least \(0.25^\circ\) or a variable-resolution mesh for a distinct and continuous depiction of the currents (Fig. 9);
-
sea-ice concentration in the Barents Sea is better represented at \(0.25^\circ\) ocean resolution compared to \(1^\circ\) with HadGEM3 and ECMWF-IFS (Fig. 10);
-
these key results are also valid when using control runs instead of historical runs.
Arctic sea ice and Atlantic OHT
The results from Sects. 3.1 and 3.2 show that: (1) in HadGEM3, ECMWF-IFS and AWI-CM, enhanced ocean resolution leads to increased poleward Atlantic OHT and decreased Arctic sea-ice area and volume; (2) in CMCC-CM2, a finer atmosphere resolution results in higher Atlantic OHT from \(20^\circ \hbox {S}\) to \(60^\circ \hbox {N}\) and slightly lower OHT from \(60^\circ \hbox {N}\) to the North Pole, as well as higher sea-ice area and volume; (3) in MPI-ESM, enhanced atmosphere resolution leads to decreased poleward OHT in the North Atlantic, increased Arctic sea-ice area and reduced sea-ice volume; (4) in HadGEM3, higher atmosphere resolution results in enhanced OHT and lower sea-ice area and volume; (5) in ECMWF-IFS, the impact of atmosphere resolution on OHT is low, while sea-ice area and volume increase with finer atmosphere resolution. In this section, we further analyze the links between Arctic sea ice (area and volume) and poleward Atlantic OHT.
The relationship between monthly mean Arctic sea-ice area and annual mean North Atlantic OHT is investigated by regressing the first variable against the second one for each month over 1950–2014. In order to isolate the relationships associated with interannual variability, we remove the trends from both variables before regressing the variables against each other. Atlantic OHT is computed at \(50^\circ \hbox {N}\), \(60^\circ \hbox {N}\) and \(70^\circ \hbox {N}\). This diagnostic thus provides the loss (or gain) in Arctic sea-ice area per PW of poleward Atlantic OHT. Figure 11 presents a synthesis of these regression slopes for 10 out of the 12 model configurations (we do not show results from ECMWF-MR and HadGEM3-HM for clarity) and each month of the year. This clearly highlights that the slopes are overall negative when OHT is computed at \(60^\circ \hbox {N}\) and \(70^\circ \hbox {N}\), meaning that sea-ice area generally decreases with increasing OHT (although the slopes can be positive for some model configurations during some months, e.g. CMCC-HR4 with OHT at \(60^\circ \hbox {N}\) in February-April). Furthermore, in general, the higher the latitude to compute OHT, the more negative the slopes, meaning that on average there is a greater change in sea-ice area per unit change in OHT (except for MPI-ESM in summer-autumn). This is especially obvious for HadGEM3 and ECMWF-IFS, with a loss in sea-ice area per PW of OHT at \(70^\circ \hbox {N}\) of \(\sim\)5-10 million \(\hbox {km}^2\) for HadGEM3-LL (Fig. 11a), \(\sim\)10-15 million \(\hbox {km}^2\) for HadGEM3-MM (Fig. 11b), \(\sim\)30-40 million \(\hbox {km}^2\) for ECMWF-LR (Fig. 11c) and \(\sim\)20 million \(\hbox {km}^2\) for ECMWF-HR (Fig. 11d).
Similar conclusions are generally drawn when we regress monthly mean Arctic sea-ice volume (instead of area) against Atlantic OHT (Fig. 12). The loss in sea-ice volume per PW of OHT at \(70^\circ \hbox {N}\) is \(\sim\)20,000 \(\hbox {km}^3\) for HadGEM3-LL (Fig. 12a), \(\sim\)20,000–30,000 \(\hbox {km}^3\) for HadGEM3-MM (Fig. 12b), \(\sim\)150,000–200,000 \(\hbox {km}^3\) for ECMWF-LR (Fig. 12c), \(\sim\)70,000 \(\hbox {km}^3\) for ECMWF-HR (Fig. 12d), \(\sim\)20,000–40,000 \(\hbox {km}^3\) for AWI-LR (Fig. 12e), mostly not significant for AWI-HR (Fig. 12f), \(\sim\)30,000–40,000 \(\hbox {km}^3\) for CMCC-HR4 (Fig. 12g), and not significant for CMCC-VHR4 (Fig. 12h), MPI-HR (Fig. 12i) and MPI-XR (Fig. 12j), although the slopes are generally negative for AWI-HR, CMCC-VHR4 and MPI-HR. While a clear correlation between Arctic sea-ice area/volume and OHT is found, these relationships do not appear to systematically change with resolution (Figs. 11 and 12).
We also compute sea-ice area in seven specific Arctic sectors, following the methodology of Koenigk et al. (2016), and we regress these sea-ice areas against the annual mean Atlantic OHT, in order to check whether some sectors have a stronger signal than others. As the regression slope between the total Arctic sea-ice area and OHT is generally more negative in winter and when OHT is computed at \(70^\circ \hbox {N}\) compared to more southern latitudes (Fig. 11), we focus on the relationship between March sea-ice area and OHT at \(70^\circ \hbox {N}\). Results from the ECMWF-HR configuration are presented in Fig. 13 to illustrate the analysis, with each dot representing March sea-ice area against annual mean OHT at \(70^\circ \hbox {N}\) for each year of the period 1950–2014. A clear anticorrelation between March sea-ice area and OHT at \(70^\circ \hbox {N}\) is found for the total Arctic (Fig. 13a), Barents/Kara Seas (Fig. 13f) and GIN Seas (Fig. 13h), with sea-ice losses of 18.3 million \(\hbox {km}^2~\hbox {PW}^{-1}\), 5.6 million \(\hbox {km}^2~\hbox {PW}^{-1}\) and 8.2 million \(\hbox {km}^2~\hbox {PW}^{-1}\), respectively. This indicates a strong potential influence of the North Atlantic OHT on the total Arctic sea-ice area, and especially in the regions closely connected to the North Atlantic (i.e. Barents/Kara and GIN Seas).
The previous result is generally valid for all other model configurations. Figure 14 shows the regression slopes of March Arctic sea-ice area against annual mean OHT computed at \(70^\circ \hbox {N}\) for the different Arctic regions and all model configurations. This shows that the area-OHT regression slopes are significantly negative in all model configurations for the total Arctic (Fig. 14a, except AWI-LR and CMCC-HR4), Barents/Kara Seas (Fig. 14f, except MPI-XR) and GIN Seas (Fig. 14h, except CMCC-HR4). For the total Arctic and these two specific sectors, no clear impact of resolution is found, beside the fact that the sensitivity of Arctic sea-ice area to OHT is slightly higher at finer ocean resolution for HadGEM3 and AWI-CM. For the other Arctic regions, results are more contrasted, with no consensus across models and much lower regression slopes. If we regress sea-ice area against OHT at \(50^\circ \hbox {N}\), no clear correlation is found for any of the Arctic regions. With OHT computed at \(60^\circ \hbox {N}\), the anticorrelation between sea-ice area and OHT is stronger for the total Arctic, GIN Seas and Labrador Sea/Baffin Bay.
The results of this section are based on historical runs (hist-1950). The use of control runs (control-1950) leads to relatively similar results, which confirms our findings related to the relationships between Arctic sea ice and OHT. In particular, area-OHT and volume-OHT regression slopes are generally negative. Also, the higher the latitude to compute OHT, the more negative the slopes, with an order of magnitude similar to historical runs. Finally, for control runs, we find no clear impact of model resolution on the intensity of the relationship between Arctic sea-ice area/volume and Atlantic OHT, similarly to historical runs. Therefore, the agreement between hist-1950 and control-1950 runs allows to rule out the possibility that Arctic sea-ice area/volume and OHT are correlated due to anthropogenic global warming. In that way, control-1950 runs allow highlighting that a mechanistic link exists between the two variables.
In the models studied, the main results related to the links between Arctic sea ice and North Atlantic OHT are summarized below:
-
regression slopes of sea-ice area/volume against poleward OHT are generally negative when OHT is computed at \(60^\circ \hbox {N}\) and \(70^\circ \hbox {N}\), meaning that an increase in OHT leads to a loss of sea-ice area and volume (Figs. 11, 12);
-
the higher the latitude to compute OHT, the more negative the regression slopes, suggesting a stronger connection between the northern North Atlantic OHT and Arctic sea ice (Figs. 11, 12);
-
area-OHT regression slopes are overall significantly negative in all model configurations for the total Arctic, Barents/Kara Seas and GIN Seas when OHT is computed at \(70^\circ \hbox {N}\) (Figs. 13, 14), suggesting these are regions where the sea ice-OHT connection is strongest;
-
these key results are also valid when using control runs instead of historical runs, highlighting that a mechanistic link exists between Arctic sea-ice area/volume and OHT.