In this paper we studied the impact of winds on Arctic sea ice through the ocean's memory by using numerical simulations. We found that the changes in halosteric height induced by wind perturbations can significantly affect the Arctic sea ice drift, thickness, concentration and deformation rates regionally even years after the wind perturbations. Changes in the Arctic liquid freshwater content and thus in halosteric height can cause changes in the sea surface height and surface geostrophic currents, which further enforce a lasting and strong impact on sea ice. The changes in both sea surface height gradient force (due to changes in sea surface height) and ice–ocean stress (due to changes in surface geostrophic currents) are found to be important in determining the overall ocean effects. The revealed ocean effects are mainly associated with changes in sea ice dynamics, not thermodynamics. Depending on the preceding atmospheric mode driving the ocean, the ocean's memory of the wind forcing can lead to changes in Arctic sea ice characteristics with very different spatial patterns. We obtained these spatial patterns associated with Arctic Oscillation, Arctic Dipole Anomaly and Beaufort High modes through dedicated numerical simulations. The dynamical impact of the ocean has strong seasonal variations, stronger in summer and weaker in winter and spring. This implies that declining trends of Arctic sea ice will very possibly allow a stronger ocean impact on the sea ice in a warming climate.
Arctic sea ice has undergone significant changes over the period of satellite observations. Not only the Arctic sea ice coverage but also the Arctic sea ice thickness have declined dramatically
Arctic geographic features and schematic of Arctic sea ice circulation (the anticyclonic Beaufort Gyre circulation over the Canada Basin, the Transpolar Drift across the central Arctic and the sea ice export through the Fram Strait, shown as blue arrows). The background gray color shows bottom bathymetry.
The decline in Arctic sea ice thickness and extent is accompanied by pronounced variability on different timescales, with contributions from both dynamic and thermodynamic processes
The Arctic liquid freshwater content varies on a quasi-decadal timescale as a memory of wind forcing
The ocean surface current and sea ice drift are strongly coupled through the ice–ocean stress
With continuing climate change there is an increasing need to predict Arctic sea ice conditions on a variety of temporal and spatial scales
In this paper we will use high-resolution numerical simulations with a global sea ice–ocean model to investigate the dynamical impact of the ocean on Arctic sea ice drift, thickness and concentration. We carried out both control and sensitivity experiments, which are the same except for the initial ocean states that have different sea surface height and surface geostrophic currents in the Arctic Ocean. The initial ocean states of the sensitivity simulations were obtained by applying wind perturbations representing different Arctic atmospheric circulation modes beforehand. By comparing the sensitivity simulations with the control simulation, we identified the impact of ocean states and different preceding wind forcings on sea ice.
The method and model setups are described in Sect. 2, and results are presented in Sect. 3. Discussion and conclusions are provided in Sects. 4 and 5, respectively.
We used the global Finite Element Sea-ice Ocean Model
We employed the multi-resolution model grid that has been used in
A control simulation was performed from 1958 to 2019 using atmospheric forcing from the JRA55-do data set
To prepare sensitivity simulations, we first carried out six simulations with wind perturbations added to wind forcing for the calculation of wind stress. They were performed for 6 years from 2010 to 2015 starting from the control run results. Using specifically designed wind perturbations representing Arctic major atmospheric modes (see the description below), the Arctic liquid freshwater content and thus sea surface height and surface geostrophic currents can be accordingly perturbed
The wind perturbations were designed as described below. The first two empirical orthogonal functions (EOFs) of deseasonalized (mean seasonal cycle removed) monthly sea level pressure (SLP) north of 70
In order to disentangle the role of the dynamical forcing imposed by the ocean on sea ice, that is, ocean–ice stress and pressure gradient force due to sea surface tilt, we carried out a few additional experiments. They are the same as the sensitivity experiments described above, except that the sea surface height in the pressure gradient force term of the sea ice momentum equation is replaced with that saved from the control simulation (see details in Sect. 3.3). One more simulation was performed to compare the dynamic impact of the ocean on sea ice with the impact of the initial sea ice state. It is the same as the sensitivity simulation with prior negative AO forcing described above, but with a different initial sea ice state (see details in Sect. 4.1).
Anomaly of halosteric height (cm) relative to the control run averaged over the 4 model years of the sensitivity simulations in which wind perturbations were switched off: experiments with an initial ocean spun up with the
Anomaly of simulated sea surface height (SSH, patch color) and surface geostrophic current (arrows) relative to the control run averaged over the four model years of the sensitivity simulations in which wind perturbations were switched off: experiments with an initial ocean spun up with the
By applying the wind perturbations, the magnitudes and spatial patterns of Arctic halosteric height and sea surface height (the dynamical sea level simulated by the model) were changed, as shown by their anomalies relative to the control run in the last (sixth) year of the wind-perturbation simulations (Figs. S5 and S6) and in the sensitivity simulations after the wind perturbations were switched off (Figs. 3 and 4). The changes in sea surface height can be explained by the changes in halosteric height, which are associated with the changes in liquid freshwater content
Different atmospheric modes can lead to changes in the halosteric height and sea surface height with very different spatial patterns (Figs. S5, S6, 3 and 4). With the negative AO perturbation, the sea surface height increases in the central Arctic, including most parts of the Eurasian and Makarov basins, while it drops in the surrounding area (Figs. S6a and 4a). With the positive AO perturbation, opposite changes are found (Figs. S6b and 4b). With the DA forcing, the magnitudes of the halosteric height and sea surface height anomalies are smaller than with the other perturbations. The negative DA perturbation increases the sea surface height in the eastern Eurasian Basin and along the southwestern periphery of the Amerasian Basin (Figs. S6c and 4c), and the changes induced by the positive DA perturbation are roughly opposite (Figs. S6d and 4d). With the BH forcing, the sea surface height changes oppositely between the Canada Basin and Eurasian Basin (Figs. S6e and f and 4e and f). A positive BH anomaly increases sea surface height in the Canada Basin and reduces it in the Eurasian Basin.
Because the wind perturbations were turned off during the sensitivity simulations, the magnitudes of the anomalies of halosteric height and sea surface height relative to the control run decreased in most of the areas in the sensitivity simulations, but the spatial patterns of the anomalies largely remained (compare Figs. 3 and 4 with Figs. S5 and S6). Therefore, the ocean kept a memory of the wind perturbations applied before.
The time series of sea surface height anomalies averaged in specific regions possessing typical ocean changes induced by the wind perturbations are shown in Fig. 5a–c. They depict the evolution of the sea surface height during the perturbation simulations (the first 6 years) and afterwards in the sensitivity simulations (the last 4 years). Consistent with the spatial patterns of the anomalies (Fig. 4), the changes in area-mean sea surface height are quasi-symmetric between the negative and positive perturbations for different forcing cases (Fig. 5a–c). Differences in the magnitudes of the sea surface height anomalies between opposite forcing phases are present regionally. For example, the magnitudes of the sea surface height anomalies are higher in the positive than in the negative AO forcing cases (Fig. 4a and b). The magnitudes of the sea surface height anomalies decrease with time in the sensitivity simulations, but they are still relatively large at the end of the sensitivity simulations, especially in the cases with initial ocean obtained with AO and BH forcing (Fig. 5a–c). The sea surface height anomalies have negligible seasonal variation in all the sensitivity simulations (Fig. 5d).
As the only difference between the settings of the sensitivity simulations and the control run is the ocean initial conditions obtained by applying wind perturbations beforehand, the anomalies of the sensitivity simulations relative to the control run can be attributed to the impact of prior wind perturbations through the ocean's memory.
We found that significant changes are induced in sea ice drift by the perturbed ocean (Fig. 6). The sea ice drift anomalies are largely aligned and scaled with the anomalies of surface geostrophic currents. In the case of prior negative AO perturbation, the sea ice drift anomalies are anticyclonic around the Arctic basin with magnitudes of 1.5–3 km/d (Fig. 6a). In the positive AO case, the sea ice drift anomalies are cyclonic (Fig. 6b). The magnitudes of the drift anomalies are larger in the case of prior positive AO perturbation, as the sea surface height and surface geostrophic currents imply (Fig. 4a and b). In the cases of prior DA perturbations, the magnitudes of sea ice drift anomalies are smaller than in other cases, and the anomalies are also less regular in space (Fig. 6c and d), as expected from the sea surface height anomalies (Fig. 4c and d). In the Eurasian Basin, the sea ice drift anomalies are anticyclonic (cyclonic) as the fingerprints of the prior negative (positive) DA forcing. In the cases of prior BH perturbations, the largest sea ice drift anomalies are found in the Amerasian Basin (Fig. 6e and f). The anomalies are anticyclonic (cyclonic) around the Beaufort Gyre in the case of prior positive (negative) BH forcing with magnitudes of about 1–2 km/d.
The significance of the impact on sea ice drift can be judged by comparing the anomalies with the mean values and the variability of sea ice drift in the control run (Fig. 6, top row). In the Beaufort Gyre region, the sea ice drift anomalies induced by the ocean states from prior BH perturbations amount to about 25 % of the mean sea ice drift in the control run. The sea ice drift anomalies in the case of prior AO forcing can even have magnitudes locally similar to those of the mean sea ice drift, although their directions are mostly not the same. Furthermore, the sea ice drift anomalies induced by the perturbed ocean are regionally much larger than the standard deviation of both annual and 5-year-mean sea ice drift.
The perturbed ocean also causes profound changes in sea ice thickness (Fig. 7). The induced sea ice thickness anomalies have different spatial patterns and magnitudes in different cases. In the case of prior negative AO perturbation, we found positive sea ice thickness anomalies from north of the Canadian Arctic Archipelago to the eastern Canada Basin with magnitudes up to 20–30 cm and negative anomalies in the western Eurasian Basin and central Arctic with magnitudes up to 10–15 cm (Fig. 7a). In the case of prior positive AO perturbation, the sea ice thickness anomalies are opposite, with larger magnitudes (Fig. 7b) in accordance with a stronger ocean perturbation (Fig. 4b). The ocean perturbed with DA forcing causes opposite changes in sea ice thickness between the Eurasian Basin and north of the Canadian Arctic Archipelago, with magnitudes up to about 10 cm (Fig. 7c and d), which are less pronounced than in other forcing cases. In the case of negative DA perturbation, the sea ice thickness anomalies are negative in the Eurasian Basin and positive north of the Canadian Arctic Archipelago. In the cases of prior BH perturbations, the largest sea ice thickness anomalies are found in the Canada Basin with magnitudes up to 15–20 cm and in the western Eurasian Basin with magnitudes up to about 10–15 cm (Fig. 7e and f). With the ocean perturbed with the positive BH perturbation, the sea ice thickness anomaly is negative in the Canada Basin and positive in the western Eurasian Basin.
The sea ice thickness anomalies induced by the perturbed ocean regionally can reach about 10 % of the mean sea ice thickness in the control run (Fig. 7, top left panel), for example, north of the Canadian Arctic Archipelago in the cases of prior AO perturbations and in the Canada Basin in the cases of prior BH perturbations. In these places, the magnitudes of the anomalies in the corresponding AO and BH forcing cases are comparable with the standard deviation of 5-year-mean sea ice thickness of the control run and slightly smaller than the standard deviation of annual mean sea ice thickness (Fig. 7, middle and right panels in the top row). Therefore, these sea ice thickness anomalies are significant considering that they persist for years following the ocean anomalies. The sea ice thickness anomalies in the cases of prior DA forcing are significant only in very small coastal areas adjacent to the Canadian Arctic Archipelago and Greenland.
We found that the influence of the perturbed ocean states on sea ice concentration is only pronounced in summer. The anomalies of September sea ice concentration in the first year of the sensitivity simulations relative to the control run are shown in Fig. 8. They indicate that the perturbed ocean can change sea ice concentration by more than 50 % regionally and shift the location of sea ice edge in the southern and western Amerasian Basin, most strongly only in the cases of prior AO and BH forcing. In the case of prior negative AO perturbation, the sea ice edge in the western Amerasian Basin is located further to the west compared to the control run (Fig. 8a), consistent with the anticyclonic sea ice drift anomalies in this region (Fig. 6a). In the opposite case with prior positive AO perturbation, the sea ice edge retreats northeastward in the southwestern Amerasian Basin (Fig. 8b), which is consistent with the cyclonic sea ice drift anomalies there (Fig. 6b). The ocean perturbed by positive BH perturbation causes sea ice edge to retreat northward in the southwestern Amerasian Basin and to slightly expand westward in the western Amerasian Basin (Fig. 8e), consistent with the anticyclonic sea ice drift anomalies in the Beaufort Gyre (Fig. 6e). The opposite impact is found in the case of prior negative BH perturbation, although the strength is weaker (Fig. 8f). The sea ice concentration anomalies regionally can be larger than the magnitudes of the variability in the control run (Fig. 8, top row), but they are confined to small areas close to ice edge. Because the impact on sea ice concentration is only significant in summer and close to sea ice edges, whose locations vary strongly in time, averaging the sea ice concentration anomalies over the 4 model years would mask the ocean impact. Therefore, in Fig. 8 we showed the September sea ice concentration anomaly in 1 particular year.
The impact of the perturbed ocean on the Arctic total September sea ice extent is not large (figure not shown). The largest impact is in the cases of prior AO perturbations, with an increase of only about 3 % for negative AO and a reduction of about 6 % for positive AO, which are associated with the changes in the ice edge locations (Fig. 8a and b). In the case of prior positive BH perturbation, the Arctic total sea ice extent is reduced by about 3 %, while the change is negligible in the case of prior negative BH perturbation because the induced negative and positive sea ice concentration anomalies along the sea ice edge nearly compensate for each other (Fig. 8e and f). The impact on sea ice concentration and sea ice edge locations weakens when the ocean anomalies weaken, as depicted by the September sea ice concentration anomalies in the last year of the sensitivity simulations (Fig. S7). In addition, the natural interannual variability of sea ice conditions could also influence the strength of the ocean effects. For example, the September sea ice concentration in the western Arctic is lower in the first model year of the sensitivity run (2016) than in the last year (2019). The less compact sea ice in 2016 potentially allows stronger impacts from the ocean.
Monthly root-mean-square (rms) difference of
The monthly root-mean-square (rms) differences of sea ice drift, thickness and concentration between the sensitivity and control simulations are shown in Fig. 9. The impact of the ocean on sea ice has strong seasonal variations. The impact is the largest in September and October, drops quickly in November and December, and stays at a relatively low level from January to May. The rms difference of sea ice drift averaged over the Arctic Ocean is in the range of 0.5–1 km/d in winter and spring and 1.5–2 km/d in summer in different cases (Fig. 9a), and the rms difference of sea ice thickness is in the range of 7–15 cm in winter and spring and 15–30 cm in summer (Fig. 9b). Compared with the seasonal Arctic mean sea ice drift between about 3 and 6 km/d and sea ice thickness between about 1 and 2 m, the rms differences are notable in all seasons. The rms sea ice concentration is very small (
The seasonal variability of the anomaly in sea surface height is negligible (Fig. 5d), so the seasonal variation in the impact is associated with the seasonal changes in sea ice compactness. Sea ice internal stress changes exponentially with sea ice concentration. At high sea ice concentration, it can partially mask the impact of the ocean (ocean–ice stress and ocean surface gradient) on sea ice momentum balance and sea ice drift. The impact of sea ice thermodynamics associated with the perturbed ocean is relatively small compared with the dynamical impact of the perturbed ocean (see the next section), so the seasonal variations in sea ice thickness and concentration anomalies are largely in phase with that of sea ice drift (Fig. 9).
There are also interannual changes in the strength of the ocean impact (Fig. 9), despite the fact that the ocean anomalies decrease with time without strong interannual variability (Fig. 5a–c). Therefore, the interannual changes in sea ice states (such as sea ice concentration, thickness and edge location influenced by atmospheric forcings) can influence the significance of the ocean impact, as the seasonality of sea ice states does.
Within the first month of the sensitivity simulations, sea ice drift has adjusted to the perturbed ocean states with rms anomalies reaching typical wintertime levels (Fig. 9a), while the rms anomalies of sea ice thickness increased with time until reaching their first maximum in the first summer of the sensitivity simulations (Fig. 9b). The anomalies of September sea ice thickness in the first year of the sensitivity simulations relative to the control run are shown in Fig. 10a–c. For brevity, we only show results from one phase of each forcing case. These anomalies indicate the total effects of the perturbed ocean from January to September, because the initial sea ice states are the same in the sensitivity and control simulations. They have spatial patterns similar to the mean anomalies averaged over the 4 model years (Fig. 7), but with magnitudes twice as large.
The differences of the changes in sea ice thickness from January to September in the first year due to thermodynamics (melting and freezing) between the sensitivity and control simulations are shown in Fig. 10d–f. Thermodynamic sea ice changes can be attributed to different processes. For example, moving more sea ice to a warmer region could enhance sea ice melting. Furthermore, changes in sea ice thickness caused by sea ice drift can change heat conduction and thus sea ice heat budget and sea ice thermodynamics
The above analysis tells us that the sea ice thickness anomalies are mainly induced by the direct sea ice dynamic effect in association with the changes in sea ice drift. In the following we will look into different ocean forcings that can influence sea ice drift and disentangle their impact on sea ice thickness. The ocean directly influences the sea ice drift through two forcing terms as shown by the sea ice momentum equation:
The anomalies of sea ice drift are largely aligned with the anomalies of ocean surface geostrophic currents (Fig. 6), which indicates that the perturbed ocean may influence sea ice drift mainly through ice–ocean stress. Does the sea surface height gradient force play an important role in sea ice drift and thickness then? To answer this question, we carried out three additional experiments. They are the same as the original sensitivity simulations with prior wind perturbations of negative AO forcing, negative DA forcing and positive BH forcing but with the sea surface height
The sea ice thickness anomalies relative to the control run for the original sensitivity simulations and their counterpart experiments with the sea surface height gradient force taken from the control run are shown in the first and second rows of Fig. 11, respectively. It is interesting to see that the sea ice thickness anomalies are quite different between the two sets of experiments. Without the perturbations in sea surface height gradient force, strong positive sea ice thickness anomalies are found in a large central area of the Arctic basin in the case of prior negative AO perturbation (Fig. 11b), in contrast to the original sensitivity experiment in which positive anomalies are mainly located from north of the Canadian Arctic Archipelago to the eastern Canada Basin (Fig. 11a). Outstanding changes are also found in the case with prior positive BH forcing. Instead of negative sea ice thickness anomalies in the Canada Basin (Fig. 11g), perturbing the ice–ocean stress alone causes positive anomalies there (Fig. 11h). In the case with prior negative DA perturbation, the positive sea ice thickness anomalies have a larger spatial extent without the perturbation in sea surface height gradient force (Fig. 11d and e). In all the three cases, eliminating the effect of sea surface height gradient changes the spatial patterns with negative and positive anomalies to patterns dominated by positive anomalies.
The differences in sea ice drift between the two sets of experiments are shown in the bottom row of Fig. 11. The drift differences (experiments with perturbations in sea surface height gradient force minus experiments without those perturbations) are orientated roughly from high sea surface height to low sea surface height (compare with Fig. 4a, c and e). Their directions and associated sea ice advections are consistent with the differences in sea ice thickness anomalies between the two sets of simulations. In the case with prior negative AO perturbation, those drift anomalies directing from the central Arctic towards the Canada Basin (Fig. 11c) can partially explain the reduced sea ice thickness in the central Arctic and increased thickness north of the Canadian Arctic Archipelago and in the Canada Basin (changing from the result in Fig. 11b to that in Fig. 11a). In the case of prior positive BH perturbation, the divergent drift anomalies in the Canada Basin (Fig. 11i) are consistent with the reduction in sea ice thickness in the Beaufort Gyre (changing from the result in Fig. 11h to that in Fig. 11g). In the case of prior negative DA perturbation, the drift anomalies directing from the eastern Eurasian Basin to the western Amerasian Basin and from the western Amerasian Basin toward the Canadian Arctic Archipelago (Fig. 11f) are consistent with the spatial contraction of positive sea ice thickness anomalies (changing from the result in Fig. 11e to that in Fig. 11d).
It is known that the sea ice momentum is mainly balanced between wind stress, ice–ocean stress and sea ice internal force
We found that the sea ice volume exports through Fram Strait in all sensitivity simulations are not significantly changed relative to the control run (see further discussion below in Sect. 4.4). Therefore, the predominant positive sea ice thickness anomalies in the simulations without perturbations in the sea surface height gradient force (second row of Fig. 11) imply that changes in sea ice thermodynamics have occurred. Indeed, the changes in sea ice thickness due to thermodynamics are predominantly positive and much stronger in these simulations than in the original sensitivity simulations (see Figs. S9 and 10d–f). That is, the impacts of ice–ocean stress alone on sea ice drift would cause overall increases in sea ice thickness in the three considered cases through both dynamic and thermodynamic changes, while the drift anomalies induced by both forcing terms together cause redistribution of sea ice and do not incur significant thermodynamic changes in sea ice. We found that influence of the ocean on sea ice concentration and edge locations also becomes different in the absence of the perturbation in sea surface height gradient force.
In summary, both forcing terms associated with the ocean in the sea ice momentum equation are important in determining the exact sea ice drift anomalies and the associated impact on sea ice thickness and concentration. Both the mean anomalies and the seasonal variation in the magnitudes of the anomalies in sea ice thickness are mainly caused by the changes in sea ice dynamics associated with the changes in sea ice drift induced by the two forcing terms together. The induced thermodynamic changes in sea ice thickness are relatively small.
Impacts of sea ice initial condition versus ocean initial condition.
Anomaly of sea ice thickness in the sensitivity experiment with only the initial ocean perturbed by applying the negative phase of Arctic Oscillation (AO) perturbation beforehand: in
In our sensitivity simulations the employed initial sea ice conditions are the same as in the control run, which allows us to easily identify the impact of the ocean memory of prior wind perturbations on sea ice. Actually, the wind perturbations during the model spinup created quite different sea ice states. If we employ sea ice states produced from the wind-perturbation simulations as initial sea ice conditions in the sensitivity simulations, can we still observe the impact of the perturbed ocean states?
We performed one additional experiment, which is the same as the sensitivity simulation with prior negative AO perturbation, but the initial sea ice state is taken from the corresponding wind-perturbation simulation rather than from the control run. The difference of initial sea ice thickness employed in these two experiments is shown at the top of Fig. 12.
The annual mean sea ice thickness anomalies relative to the control run for each year in these two simulations are shown in Fig. 12a–d and 12e–h, respectively. We see that the anomalies of the two simulations become more similar with time, indicating that the dynamic impact of the ocean becomes more important than the initial perturbations in sea ice. Even with this very large initial difference in sea ice (up to about 1 m in thickness), the sea ice thickness anomalies relative to the control run are already dominated by the impact of the perturbed ocean after 2 years and fully represent the ocean impact in the fourth year. This simulation further proves the robustness of our findings on the impact of the ocean's memory. We note that there could be effects from initial sea ice states through potential feedbacks in coupled climate models, which is beyond the scope of this study. Here we used forced sea ice–ocean simulations to reveal the direct impact of the ocean on sea ice.
Changes in sea ice drift, thickness and concentration are associated with changes in sea ice deformation. The deformation rate of pack ice is a key parameter determining the formation of sea ice linear kinematic features like leads
In the control run, larger deformation rates are located in the Beaufort Gyre region and in the marginal seas (top row of Fig. 13), consistent with observations
Our results reveal that the impact of the ocean anomalies on sea ice deformation can be relatively strong and suggest that understanding changes in sea ice deformation needs to consider both the atmosphere and ocean forcings. Dedicated studies on the impacts of the ocean on sea ice deformation rates are still required in the future.
In this study we used idealized wind perturbations representative of major Arctic atmospheric modes to spin up the ocean for 6 years before the 4-year-long sensitivity simulations. The length of the simulations is within the observed range of typical variation periods of Arctic freshwater content. The amount of liquid freshwater in the Arctic Ocean varied at 5- to 7-year intervals before 1997 and has remained in a state of accumulation for about 2 decades since then
Over the last 2 decades the Arctic Ocean has accumulated about 11 000 km
We found that the impact of the perturbed ocean on sea ice has strong seasonal variations, which is due to the seasonal variation in sea ice internal stresses. The implication is that the impact of the ocean on sea ice will become stronger in a warming climate as sea ice will become thinner and less compact. It was also found that winds can perturb ocean freshwater content, sea surface height and geostrophic currents more strongly with sea ice decline between the recent sea ice condition and that in the 1980s
In our simulations the ocean memory of prior wind perturbations influences sea ice mainly dynamically through the anomalies of sea surface height and surface geostrophic currents. The upper ocean temperature changes caused by winds are small and have much smaller contributions to the obtained sea ice anomalies than the dynamic impact. This is shown by the small sea ice thickness changes associated with sea ice thermodynamics (Fig. 10). Furthermore, the dynamic impact can reemerge even though it is masked by applying large sea ice initial perturbations at the beginning (Fig. 12). The implication is, even if there are additional initial thermal perturbations that can induce initial sea ice anomalies, the ocean dynamic effect can reemerge as long as the perturbation signals in sea surface height and surface geostrophic currents exist.
We analyzed the dynamic impact of the ocean on sea ice on monthly and longer timescales in this study. As the strength of the impact is very sensitive to the temporal changes in sea ice as revealed by the strong seasonal variability of the impact, regional variability of sea ice internal stress on shorter timescales could also allow for high-frequency variability in the impact of the ocean on sea ice drift. However, on short timescales wind variability is strong, and it determines most of the sea ice motion
The regulation of sea ice states on the impact of the ocean implies that an overestimated sea ice thickness in our model might lead to an underestimation in the induced sea ice changes in the sensitivity experiments. We also note that the idealized wind perturbations we used were intended to allow for easy interpretations of involved dynamical processes. Realistic wind forcing can produce more complicated spatial structures in the response of sea ice.
The indirect impact of winds on sea ice through the ocean is significant as we revealed in this paper, but it is weaker than the direct impact of wind perturbations when they are present. For comparison, the differences of sea ice thickness, drift, concentration and deformation rates in simulations with the presence of wind perturbations relative to the control run are shown in Figs. 14 and S10–S13. The spatial patterns of the direct impact of winds on sea ice could be different from or similar to those induced through the ocean’s memory depending on cases and properties investigated. For example, the direct and indirect impacts of BH forcing are similar for the deformation rates in terms of the spatial patterns but quite different for the sea ice thickness. Another example is for the DA forcing, which has a very strong direct impact on sea ice thickness, concentration and area but only a weak impact through the ocean's memory because of its weak influence on halosteric height.
As mentioned above, the impact of the perturbed ocean on sea ice volume export through Fram Strait is small. The export anomalies are around 100 km
In this numerical study we used wind perturbations to change the Arctic Ocean and then used the perturbed ocean as initial conditions in sensitivity simulations to investigate the impact of wind perturbations existing beforehand on Arctic sea ice. The wind perturbations can change the sea surface height and surface geostrophic currents through changing the ocean's liquid freshwater content, an integral indicator of the upper ocean salinity and thus the halosteric height. In the sensitivity simulations the wind perturbations are turned off, so the previously induced changes in halosteric height, sea surface height and surface geostrophic currents weaken with time, but they can last for years. We found that these lasting changes in the ocean can significantly influence the sea ice drift, thickness, concentration and deformation rates, which manifests a lasting impact of winds on sea ice through the ocean's memory.
The spatial patterns of the changes in sea ice states induced by the perturbed ocean depend on the wind perturbations applied beforehand. We investigated the cases with the ocean perturbed by winds representing the Arctic Oscillation (AO), Arctic Dipole Anomaly (DA) and Beaufort High (BH) modes for both their positive and negative phases. In the case of prior negative AO perturbation, sea ice thickness is oppositely changed by the ocean perturbation in two regions. One is the north Canadian Arctic Archipelago and western Canada Basin where sea ice thickness increases, and the other is the western Eurasian Basin and central Arctic where it decreases. The changes are opposite in the case of prior positive AO perturbation. In the case of positive (negative) BH perturbation, sea ice thickness is decreased (increased) in the Canada Basin and increased (decreased) in the western Eurasian Basin. In both cases of prior AO and BH perturbations, the ocean, which was perturbed to an extent well within the observed range, can change annual mean sea ice thickness by up to about 10 % regionally in our simulations. Compared with the standard deviation of annual and 5-year-mean sea ice thickness, the persistent sea ice thickness anomalies are regionally significant.
In general, the induced sea ice drift anomalies are largely aligned and scaled with the anomalies of ocean surface geostrophic currents. The ocean perturbed with BH forcing can change annual mean sea ice drift speed by up to about 25 % in the Beaufort Gyre region, while the ocean perturbed with AO forcing can lead to even stronger relative changes in sea ice drift locally. Compared with the variability of sea ice drift, these drift anomalies are very significant.
The impact of the perturbed ocean on sea ice concentration is pronounced only in summer. The ocean perturbed by AO and BH forcing can moderately influence the location of sea ice edge regionally in the western Arctic in September, with the Arctic September sea ice extent changed by up to about 3 %–6 %. In the cases of prior DA perturbations, the weaker perturbation of the ocean has a weaker impact on sea ice compared to other cases, and the induced changes in sea ice thickness and concentration are not significant compared with the magnitudes of the interannual and pentadal variability. Our simulations further revealed that the ocean can significantly influence sea ice deformation rates.
We found strong seasonal variations in the changes of sea ice drift, thickness and concentration associated with the perturbed ocean. Although the sea ice changes are larger in summer and smaller in winter and spring, they are significant year-round for sea ice drift and thickness. Because the seasonal variations in the ocean (sea surface height and geostrophic currents) perturbations are negligible, the seasonal variations in the sea ice changes can only be explained by the seasonal variability of sea ice internal stress. Weaker internal stress allows the impact of the ocean to be more pronounced. This implies that the impact of the ocean on sea ice in the future warming climate will become stronger.
Our analysis revealed that the impact of the perturbed sea surface height and surface geostrophic currents on sea ice is mainly through changing sea ice drift and sea ice dynamics, not sea ice thermodynamics. We also found that the changes in both ice–ocean stress and sea surface height gradient force, two of the sea ice momentum budget terms, are important in determining the exact changes in sea ice. The changes in sea ice thickness would be fully different without the subtle role of sea surface height gradient force.
The strong impact of ocean dynamic changes on sea ice suggests that not only the changes in the atmosphere should be considered for understanding observed changes in sea ice, but also the changes in the ocean and the changes in the response of the ocean to the atmosphere should be taken into account. Sea ice drift pathways changed significantly between different years even for ice floes originating from the same location
The FESOM model version used in this study is available at
The supplement related to this article is available online at:
QW conceived this study, performed the simulations and wrote the first draft of the paper. All coauthors contributed to the data processing, model result interpretation and paper improvement.
The authors declare that they have no conflict of interest.
Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
The authors thank the reviewers (Céline Heuzé and Leandro Ponsoni) and the editor (Michel Tsamados) for their helpful comments. This work was supported by the German Helmholtz Climate Initiative REKLIM (Regional Climate Change), by the FRontiers in Arctic marine Monitoring program (FRAM), by the German Federal Ministry for Education and Research (BMBF) within both the GROCE2 project (grant number 03F0855A) and the EPICA project (grant number 03F0889A), and by the Collaborative Research Centre TRR 181 “Energy Transfer in Atmosphere and Ocean” funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) with project number 274762653. The simulations were carried out at North German Supercomputing Alliance (HLRN).
This research has been supported by the Deutsche Forschungsgemeinschaft (grant no. 274762653) and the German Federal Ministry for Education and Research (grant nos. 03F0889A and 03F0855A). The article processing charges for this open-access publication were covered by the Alfred Wegener Institute, Helmholtz Centre for Polar and Marine Research (AWI).
This paper was edited by Michel Tsamados and reviewed by Leandro Ponsoni and Céline Heuzé.