1. Motivation
The fast decline of Arctic sea ice became the major indicator of global climate change, with consequences for, among others, Arctic exploration and exploitation of natural resources, shipping and commercial transits (Melia et al. 2016), living habitats, ecosystem structure, and stability. The Arctic summer sea ice extent has declined by roughly a half since the 1970s. For every year since 2007, the annual minimum of sea ice extent in September has been lower than in any other year before 2007 (Stroeve and Notz 2018). The reduced ice cover becomes thinner (Lindsay and Schweiger 2015), younger (Kwok 2018), and more mobile (Rampal et al. 2011). By midcentury, the Arctic will likely become summer ice-free under business-as-usual emission scenarios according to Coupled Models Intercomparison Project (CMIP) projections (Sigmond et al. 2018; Notz and Stroeve 2016; Wang and Overland 2012). Thus, urgent action is needed to better understand Arctic sea ice predictability and to develop advanced forecast methods for all seasons at pan-Arctic and regional scales (Eicken 2013).
Community efforts in advancing capabilities to predict Arctic sea ice were consolidated by the Sea Ice Outlook (https://www.arcus.org/sipn/sea-ice-outlook), which, starting from 2008, collects and synthesizes predictions of summer sea ice extent, with now over 500 predictions based on dynamical modeling and statistical and heuristic methods. An important outcome of this initiative is an improved understanding of processes driving sea ice predictability (Hamilton and Stroeve 2016). Particularly, they found that for extreme years when sea ice extent departs far from its downward long-term trend, predictions showed lower skill; a lack of statistically significant differences for July and August absolute prediction errors suggests that the increased predictive skill in normal years is not connected to the selected lead time. Sensitivity of predictions to initial conditions such as sea ice thickness (e.g., Day et al. 2014; Blanchard-Wrigglesworth et al. 2017), melt pond fraction (Schröder et al. 2014), and timing of sea ice retreat (Stroeve et al. 2016; Petty et al. 2017) was emphasized. Oceanic conditions such as influx of heat through the Bering Strait also play an important role in the predictability of Arctic sea ice state (Serreze et al. 2016a). In that, synoptic-scale predictability of the sea ice state is limited given the known (7–10 days) limits of predictability of weather conditions (e.g., Blanchard-Wrigglesworth et al. 2011a; Serreze et al. 2016b). Longer (e.g., seasonal) predictability of the mean sea ice state resides in the slowly (and more deterministic) changing ocean thermodynamic conditions and low-frequency coupled ocean–atmosphere modes of variability (e.g., Lindsay et al. 2008). Still, whether the new state of the Arctic climate system would enhance or restrain predictability of the seasonal evolution of the sea ice has remained largely unknown.
In this study, we combine results of observations, state-of-the-art reanalysis, and seasonal forecasts of Arctic sea ice to demonstrate competing influences of dynamics and thermodynamics on the change of sea ice thickness predictability from the period 1993–2006 to the period 2007–20. The assessment of changes to Arctic sea ice predictability is carried out here using an operational seasonal forecasting system, rather than idealized experiments as in Tietsche et al. (2013) or Holland et al. (2019). We start this paper with a brief introduction (section 2) describing changes relevant to sea ice state in all physical components of the Arctic climate system. Descriptions of data and methods are provided in section 3. Then, in section 4, we demonstrate the ability of the reanalysis to capture major high-latitude changes in recent decades. Next, in section 5, we discuss predictability (including the role of dynamics and thermodynamics) of the Arctic sea ice thickness over the growth-to-melt season and the change of predictability from 1993–2006 to 2007–20. A discussion (section 6) summarizes our findings, placing them in a broader context.
2. Introduction
a. Changes in Arctic atmospheric and ice dynamics
Arctic atmospheric circulation and sea ice drift have exhibited dramatic changes during past decades in conjunction with the accelerating warming of the Arctic. For example, the leading atmospheric circulation pattern associated with the large-scale Arctic Oscillation reflects the strength of the polar vortex (Thompson and Wallace 1998). In the late 1980s to early 1990s, a high state of the Arctic Oscillation contributed to a reduction of Arctic sea ice (e.g., Rigor and Wallace 2004). In the 2000s, the strengthened atmospheric high over the Amerasian Basin and low sea level pressure in the eastern Arctic during the negative (anticyclonic) Arctic Oscillation phase enhanced the surface anticyclonic Beaufort Gyre and Transpolar Drift, which together move ice and cold fresh upper ocean waters from the Siberian shelf across the central Arctic toward Fram Strait. This atmospheric circulation pattern increased atmospheric heat transport into the Arctic, reducing sea ice coverage there (e.g., Zhang et al. 2008; Graversen et al. 2011). The Arctic Oscillation alternates with another pattern called the Arctic dipole pattern, in which persistently high pressure is observed over the Beaufort Sea and low pressure over the Kara and Laptev Seas (e.g., Wu et al. 2006). Both these patterns resulted in a poleward shift in storm tracks and an intensification of storm activity in recent years and were suggested as important drivers in sea ice reduction (e.g., Zhang et al. 2004; Serreze and Barrett 2008; Sorteberg and Walsh 2008; Simmonds and Keay 2009).
However, in the 2010s, the Arctic Oscillation underwent a drastic spatiotemporal shift. Since 2014–15, the sea level pressure over the Beaufort–Chukchi region was steadily decreasing in winter months (Ballinger et al. 2021), with a “collapse” of the Beaufort high in winter 2016–17 (Moore et al. 2018; Babb et al. 2020). This collapse was characterized by the lowest western Arctic sea level pressure since 1979 and associated with winter-long reversal of the surface winds and sea ice drift (Moore et al. 2018). This tendency toward lower sea level pressure in the western Arctic culminated in 2020 when the Arctic Oscillation index reached record-high value of 2.83, with remarkable persistence of the extremely anomalous Arctic Oscillation over all winter months (Ballinger et al. 2021). Winter 2020 was also characterized by unusually frequent storms and anomalous westerly winds driving an eastward ice drift across the Amerasian Basin toward Fram Strait (Ballinger et al. 2021). Attenuation of the Beaufort Gyre increases the residence time of ice in the region and, consequently, ice accretion due to growth and deformation (Babb et al. 2020). As we will see later, the changes in sea ice dynamics in recent years altered predictive skills of the seasonal sea ice evolution forecasts.
b. Contrasting regional oceanic changes
Observations in the 2010s have captured unprecedented changes in the Eurasian Basin (EB) of the Arctic Ocean, likely representing a fundamental shift to a new, more dynamic environment (for geographical names, see Fig. 1). For example, in 2013–18 a notable warming and salinification compared to the climatological mean were found in the halocline (60–150 m), a layer of cold fresher water isolating the heat associated with the intruding underlying (150–900 m) warm (temperature > 0°C) and salty Atlantic Water (AW) from the surface mixed layer (SML) and from sea ice. The combination of weaker and thinner halocline and shoaling of the AW, coupled with net loss in ice volume, has allowed progressively deeper winter ventilation in the eastern EB (Polyakov et al. 2017, 2020a). This ventilation has resulted in enhanced upward AW heat fluxes, which were sufficiently large to contribute substantially to the recently diminished sea ice cover (Polyakov et al. 2017, 2020a). In that, the role of oceanic heat becomes dominant (compared with atmospheric contribution) in melting EB sea ice (Polyakov et al. 2020a), potentially leading to enhanced predictability of seasonal thermodynamic sea ice changes in this region.
Arctic Ocean map with three regions identified: the Eurasian Basin subregion (EBe; blue), abyssal Amerasian Basin (ABa; light red), and slope Amerasian Basin (ABs; dark red) are indicated (ABa and ABs are separated using a value of available potential energy; the corresponding isoline is shown in Fig. 2f). The Lomonosov Ridge (LR), Novosibirskiye Islands (NI), Severnaya Zemlya (SZ), Franz Joseph Land (FJL), Svalbard (SV), Makarov Basin (MB), and Fram Strait (FS) are also indicated. Pathways of intermediate Atlantic Water (AW) and Pacific Water (PW) are shown by red and purple arrows.
Citation: Journal of Climate 35, 9; 10.1175/JCLI-D-21-0463.1
Note that the release of heat from the ocean interior is very seasonal and is related to the winter sea ice formation associated with salt expelled from ice into the water, as evidenced by observational (e.g., Polyakov et al. 2017, 2020a; Pérez‐Hernández et al. 2019) and model-based studies (e.g., Ivanov et al. 2018; Athanase et al. 2020). This brine-enriched water is heavier and sinks (this mixing process is called winter convection). In the absence of a strong halocline, the cold salty water mixes much more efficiently with the AW. As a result, this AW heat is then transferred upward to the bottom of sea ice, limiting the amount of ice that can form during winter. The maximum depth of the winter convection is reached in late April–early May. Winter decrease of the rate of sea ice formation due to oceanic heat should have a direct impact on the state of the sea ice cover through the summer melting season. That provides a predictive scale for the seasonal thermodynamically driven sea ice evolution of several months.
By contrast, in the Amerasian side of the Arctic enhanced freshening and deepening of the surface fresh layer due to local processes like sea ice melt, redirection of Siberian riverine waters into the Beaufort Gyre, and intensification of Arctic high and wind-driven convergence of the upper ocean (e.g., Yamamoto-Kawai et al. 2009; Morison et al. 2012; Proshutinsky et al. 2009; McPhee et al. 2009) make the separation between the surface and deep layers more pronounced. As the pool of freshwater grows, it limits mixing and the heat transfer from the ocean interior to the surface, diminishing the role of ocean thermodynamics. On the other hand, effects of increasing influx of warm Pacific Water (PW; a key constituent of halocline in the AB) through the Bering Strait (Woodgate 2018) are well pronounced at the deep basin’s margins where PW enters the basin interior (e.g., Polyakov et al. 2020b; Danielson et al. 2020) and then spreading into the Beaufort Gyre interior (Proshutinsky et al. 2019). Shimada et al. (2006) and Serreze et al. (2016a) emphasized the important role of the PW heat for the sea ice edge position in the Chukchi Sea and beyond into the Beaufort Sea. Thus, we argue that, similar to the effect of AW, improved predictive skills should be expected in the southern AB under the impact of an anomalously strong influx of PW.
3. Data and methods
a. ORAS5 reanalysis and SEAS5 forecasts
We use data from the European Centre for Medium-Range Weather Forecasts (ECMWF) Ocean Reanalysis System 5 (ORAS5) and its near real-time extension OCEAN5 (Zuo et al. 2019). The ocean model of ORAS5 has an eddy-permitting horizontal resolution 0.25°. Vertical resolution decreases from 1 m in the very top to ∼5 m around 50-m depth to ∼20 m around 210 m, with 31 of the total 75 layers lying above 210 m. Coarser resolution in deeper layers is not problematic for simulating oceanic heat content [see, e.g., the appendix of Polyakov et al. (2003)] but may be a limiting factor for simulating oceanic heat fluxes. It includes the thermodynamic–dynamic sea ice model LIM2. ORAS5 has five ensemble members that cover the period since 1979. Atmospheric forcing of ORAS5 comes from ERA-Interim (Dee et al. 2011) up to 2016 and the ECMWF operational analysis afterward (Mayer et al. 2019). The forecast model is ECMWF’s seasonal forecast system SEAS5 (Johnson et al. 2019), which has been operational since 2017. SEAS5 forecasts consist of a 51-member ensemble (7 months lead time) and a 15-member extension (13 months lead time). Here we use forecasts initialized on 1 February. Since we are interested in the sea ice thickness evolution up to September (corresponding to 8-month-long forecasts), we mainly draw on the 15-member extension of the SEAS5 forecasts. For diagnostics on shorter time ranges we checked results using the larger ensemble to confirm their robustness. Operational forecasts are combined with historical forecasts to cover the period 1993–2020. Note that while many other studies (e.g., Bushuk et al. 2020) focus on predictability of sea ice area, we focus on predictability of sea ice thickness.
b. Estimating vertical oceanic heat fluxes
c. Sea ice budget
SEAS5 archives daily averaged effective (i.e., gridpoint-averaged) ice thickness fields, which allows to compute accurate monthly total effective thickness tendencies dH/dtTot. SEAS5 also archives monthly accumulated thermodynamic contribution to thickness changes dH/dtTh. Combining these terms allows indirect computation of the dynamic contribution dH/dtAd to monthly thickness changes via dH/dtAd = dH/dtTot − dH/dtTh.
d. Predictable component
Evaluation of predictability of seasonal evolution of the Arctic sea ice thickness is done using the predictable component (PC) of the model (Eade et al. 2014). The PC estimates the fraction of variability that is potentially predictable by computing the temporal standard deviation (across start dates) of the forecast ensemble means normalized by the spread of all individual ensemble members from all start dates:
Changes in predictability are diagnosed by comparing PCs computed separately for the two 14-yr-long early (1993–2006) and late (2007–20) subperiods of our full study period. This choice is motivated by the fact that 2007 was the first year with very low summer sea ice extent, reflecting changes toward the “new Arctic” (e.g., Carmack et al. 2015) and their potential impact on sea ice predictability.
e. Available potential energy
f. Sea ice drift vectors
We used monthly sea ice motion vectors based on the Polar Pathfinder Daily 25 km EASE-Grid Sea Ice Motion Vectors product in version 4.1 (NSIDC-0116; Tschudi et al. 2020). Input data used to generate daily and weekly sea ice motion estimates are the AVHRR, AMSRE, SMMR, SSMI, and SSMI/S satellite sensors, International Arctic Buoy Programme (IABP) buoys, and NCEP–NCAR reanalysis forecasts. Data are available online at https://icdc.cen.uni-hamburg.de/en/seaicedrift-satobs-global.html.
4. Simulated response of Arctic sea ice and upper ocean to climate change
In this study, we analyze the Arctic sea ice response to dynamic and thermodynamic forcing and predictability of the sea ice state utilizing results of state-of-the-art ocean analysis and seasonal prediction systems. First, we demonstrate that the ocean reanalysis ORAS5 captures the pattern of major high-latitude changes evident from observations (cf. Uotila et al. 2019). Second, we show that these changes lead to an alteration of regional seasonal predictability of the sea ice in comparison to the past. To demonstrate this, we use seasonal forecasts from SEAS5.
a. Change of upper ocean thermohaline state in ORAS5
According to the ORAS5 reanalysis, there is a nearly ubiquitous pattern of warming and freshening of the upper 50 m in 1993–2017 (Figs. 2a,b), which is consistent with recent observations (Tables 1 and 2; see also Polyakov et al. 2020b, their Fig. 9). For these comparisons, observational estimates are obtained following the method and using the same data as in Polyakov et al. 2018. Particularly, available potential energy (APE; see definition in section 3) is used here as an integral indicator of changes in Arctic halocline and surface mixed layer strength (Polyakov et al. 2018). These changes result from amplified atmospheric thermodynamic forcing (e.g., Carmack et al. 2015). We note that, while the pattern of surface warming and freshening is well captured by the model, the magnitude of the observed changes is significantly underestimated (Table 2). Warming in the deeper (halocline and AW) layers is attributed to the advected heat of the Atlantic and Pacific origins (e.g., Polyakov et al. 2020b). Salinity (Figs. 2c,d) and stratification (Figs. 2e–g) changes derived from reanalysis show striking regional differences. Particularly, in the EB, shoaling of the AW layer and injection of saltier and denser water from the Barents Sea into the EB halocline makes this layer of the ocean saltier (Table 2), denser and, as a result, less stratified and more susceptible to mixing. These simulated changes are highly consistent with mooring observations (e.g., Polyakov et al. 2020b; see their Figs. 7 and 8, and our Tables 1 and 2), providing strong support for the reanalysis-based results. At the same time, the reanalysis shows that the upper central AB has become more stratified (Table 1), thus constraining communication between the surface and underlying waters and limiting vertical mixing. The estimated annual trend derived from simulations is lower compared with the one from observations (Table 2), which should have no major impact on already negligible AW heat fluxes in the central AB (e.g., Carmack et al. 2015). Contrasting changes of the simulated APE trends over 1993–2017 toward weaker stratification in the upper EB are also consistent with the recent observational findings (cf. similar annual EB APE trends obtained from simulations and observations; Table 2). Note that, similar to observations, the magnitude of the simulated APE trends in the AB is greater than in the EB. It is important for the discussion here that there is a clear advective pattern of change in time of the simulated APE associated with anomalous influx of AW in the EB and PW in the AB (Fig. 2g)—a process referred to as borealization of the Arctic Ocean (Polyakov et al. 2020b). The effects of increasing influx of PW through the Bering Strait are well pronounced at the basin’s periphery where PW enters the polar basin (Fig. 2g).
ORAS5-based depth vs latitude sections of 1993–2017 trends of annual (a),(b) temperature (°C yr−1) and (c),(d) salinity (psu yr−1) in the (a),(c) Eurasian Basin (EB) and (b),(d) Amerasian Basin (AB) of the Arctic Ocean. (e) Time series of available potential energy (APE) anomaly relative to the 1993–2017 mean (J m−2) in the upper ocean (SML + halocline) averaged over the EB and AB. Dotted lines show 95% confidence intervals. Also shown are maps of (f) 1993–2017 mean APE (J m−2) and (g) APE annual linear trends (J m−2 yr−1); in (f), a solid black line identifies the boundary between ABs and ABa subregions of the AB; hatching in (g) is added where trends are not statistically significant following Student’s t test.
Citation: Journal of Climate 35, 9; 10.1175/JCLI-D-21-0463.1
Regional Arctic 1993–2017 means (plus and minus two standard errors) from ORAS5 and observations for the EB’s region EBe and AB’s abyss and slope regions ABa and ABs. The top section shows surface mixed layer (SML) means of potential temperature θ and salinity S. The bottom sections shows halocline means of potential temperature θ, salinity S, and available potential energy APE.
Regional Arctic trends in 1993–2017 from ORAS5 and observations for the EB’s region EBe and AB’s abyss and slope regions ABa and ABs. The top section shows surface mixed layer (SML) trends of potential temperature θ and salinity S. The bottom section shows halocline trends of potential temperature θ, salinity S, and available potential energy APE. Italic font indicates trends that are not statistically significant.
b. Heat fluxes in ORAS5
Yet again, the ORAS5 reanalysis shows contrasting regional changes of heat flux Fh (see section 3) from the ocean interior, which are highly consistent with observations (Carmack et al. 2015, their Table 1). Particularly, weaker halocline in the EB than in the AB fosters stronger heat exchanges between the ocean interior and the SML and sea ice so that the simulated Fh at the base of the EB winter SML (50 m) exceeds those in the AB (Figs. 3a,b,g,h; areas used for Fh estimates are indicated in Fig. 1; they are closely related to the regional spatial patterns of APE in Figs. 2f,g). Furthermore, the simulated weakening of halocline stratification in the EB and increasing influx of AW heat through Fram Strait in recent decades have altered the regional upward heat transport from the ocean interior in such a way that over the last 25 years the simulated Fh in the EB increased by ∼0.7 ± 0.5 W m−2 (estimated from the linear trend), in contrast with the lack of any change of Fh in the central AB (e.g., see regional EB and AB trends in Figs. 3a,b). This EB flux change is not negligible, considering that the excess net energy flux into the ice pack required to explain the observed decay of sea ice coverage over the several most recent decades is just ∼1 W m−2 (Kwok and Untersteiner 2011). In the AB, the simulated Fh values do not exceed 1 W m−2 in the central basin and are 2–3 W m−2 over the slope and Chukchi Cap areas (Fig. 3b).
Evolution in time of oceanic heat fluxes Fh (W m−2) in the EB’s region EBe as evidenced from by (a) Fh at 50 m, and by Hovmöller diagrams of long-term (c) seasonal means and (e) annual trends of Fh. (b),(d),(f) As in (a), (c), and (e), but for the AB’s regions ABa and ABs (areas are indicated in Fig. 1). Also shown are ORAS5-based maps of the 1993–2017 (g) mean and (h) decadal linear trends of oceanic heat flux Fh at 50-m depth [positive (red) is upward]; hatching is added where trends are not statistically significant following Student’s t test.
Citation: Journal of Climate 35, 9; 10.1175/JCLI-D-21-0463.1
Placing these reanalysis-based estimates in the perspective of existing sparse observations, we note that mean divergent heat flux of 10.4 W m−2 inferred from mooring observations in the eastern EB for five winters from 2013–14 through 2017–18 greatly (threefold) exceeded the previous estimates of 3–4 W m−2 derived from summer 2007–08 microstructure observations over the Laptev Sea slope and winter 2009–10 observations in the central Amundsen Basin (Polyakov et al. 2020a). Thus, the simulated EB Fh is within the range of regional observations, while the increase in the simulated Fh is somewhat weaker compared with observational estimates. The magnitude of the simulated Fh in the AB is highly consistent with observations staying within 0.1–0.3 W m−2 in the central basin and reaching several watts per square meter at the steep topography (see Table 1 from Carmack et al. 2015). It is essential for the goals of our study that the model captures the general tendency of increasing EB Fh and lack of trend in the central AB; considering that the magnitude of changes in the EB is underestimated by the model, the conclusions derived here should be viewed as conservative.
The reanalysis-based seasonal cycle in the upper (SML + halocline) ocean is also strikingly different in the EB and AB (Figs. 3c–f). In both basins the model simulates SML cooling associated with heat losses by the upper ocean (and positive Fh at the base of the SML) in winter and SML warming as a result of atmospheric radiative forcing (and negative Fh at the base of the SML) in summer. In the EB, however, winter ventilation is not limited by the SML as in the AB. Instead, it penetrates the deeper halocline layers reaching 200-m depth (Fig. 3c), resonating with recent mooring observations of profound halocline ventilation in the eastern EB (e.g., Polyakov et al. 2020a). Transition of the upper ocean from cooling to warming occurs in mid-April. Figures 3e and 3f show that these basic features of the upper ocean ventilation were amplified by the trend in recent decades. It is manifested as a strong increase in time of atmospheric energy pumped into the SML in summer in both basins and enhanced AW heat losses from the EB interior through the halocline in winter. This reanalysis-based finding is consistent with and provides broader geographical context to recent observations (Pérez‐Hernández et al. 2019; Polyakov et al. 2017, 2020a).
c. Sea ice drift in SEAS5
Switching now to the forecast model, SEAS5-simulated winter [February–April (FMA)] sea ice drift averaged over 1993–2018 is shown in Fig. 4a. For comparison, Fig. 4b shows FMA sea ice drift derived from 1993 to 2018 satellite observations and Fig. 4c shows their differences (see section 3 for details). The model successfully reproduces all major features of the Arctic sea ice drift, as attested by rather small differences of simulated and observed ice drift vectors that do not exceed 1 cm s−1 in the Arctic Ocean proper; greater errors are found at the Arctic Ocean periphery (Fig. 4c), which is not relevant to our analysis of Arctic sea ice predictability. Namely, the simulated climatologic pattern of sea ice drift is realistically dominated by the anticyclonic Beaufort Gyre, which is the key to control perennial ice in the western Arctic (e.g., Babb et al. 2020) and the Transpolar Drift carrying ice from the Siberian Arctic toward Fram Strait. The Pan‐Arctic pattern of ice drift converges sea ice at the northern coast of Greenland and the Canadian Arctic Archipelago (Kwok 2015), establishing the thickest sea ice in the Arctic (Bourke and Garrett 1987).
(top) Climatological February–April sea ice drift in (a) SEAS5 1993–2020 and (b) NSIDC data 1993–2018, and (c) their difference (colors depict ice speed; m s−1). (bottom) Decadal trends in February–April sea ice drift in (d) SEAS5 1993–2020 and (e) NSIDC data 1993–2018, and (f) their difference (colors depict ice speed trend; m s−1 decade−1).
Citation: Journal of Climate 35, 9; 10.1175/JCLI-D-21-0463.1
As demonstrated in Figs. 4d–f, the model underestimates trends in sea ice drift when compared to satellite-based estimates. Satellite observations reveal a pan‐Arctic increase in ice drift speeds, which Rampal et al. (2009) and Spreen et al. (2011) ascribed to the mechanical weakening of the Arctic ice pack. Particularly apparent from satellite observations is an accelerated ice drift in the southern Beaufort Sea. Overall, simulated ice drift speeds are not changed much over the last three decades in the EB; there was some nearslope intensification of ice drift in the eastern EB whereas in the western EB, contrary to observations, slight negative trends dominated the Transpolar Drift. However, the model was successful in simulating the observed acceleration of the Beaufort Gyre in recent decades. The simulated ice motion along the southern flank of the climatological Beaufort Gyre accelerated at the rate of 1–2 cm s−1 decade−1. The acceleration of westward ice drift along the Beaufort Gyre’s southern boundary decreases ice residence time in the region and promotes more rapid advection of thick ice from the area north of the Canadian Archipelago which, as we show later, is critical for the change of seasonal sea ice predictability in the region.
d. Changes in seasonal sea ice evolution in SEAS5 forecasts
We now turn to results from seasonal predictions initialized from ORAS5 on 1 February. The 1993–2020 winter (FMA) climatology from SEAS5 shows expected patterns of seasonal ice thickening (Fig. 5a), with a strong thermodynamic contribution everywhere except for the ice edge (Fig. 5b), and ice convergence at the sea ice edge and toward the Canadian Arctic Archipelago (Fig. 5c). However, analysis of trends highlights differences in the role of dynamic and thermodynamic drivers in shaping the Arctic sea ice changes at regional scales (Figs. 5d–f). The most significant changes in the rates of the simulated thermodynamic sea ice production are found in the northern Kara, Barents, and Greenland Seas (Fig. 5e). In the Arctic Ocean proper, the simulated thermodynamic sea ice changes show noticeably weaker trends, with reduction of sea ice production along the Canadian Arctic Archipelago and Nansen Basin, and increased rate of thickening elsewhere (Fig. 5e). Particularly, we emphasize a reduced thermodynamic winter ice production rate in the eastern EB—a clear sign of Atlantification—as evidenced by ∼−0.5–1.0 cm month−1 decade−1 trend, which is equivalent to a ∼10–20-cm decrease of winter sea ice growth over the last 35 years. The model underestimates the observed thermodynamic sea ice loss rates (cf. with 78–93-cm reduction of sea ice growth from the early 2000s to 2017–18; Polyakov et al. 2020a); however, even this conservative estimate represents a sizeable (∼10%) contribution to regional sea ice reduction in the eastern EB (Fig. 5b; Labe et al. 2018).
Maps of (top) climatology (1993–2020) and (bottom) linear trend of February–April seasonal ice thickness changes in SEAS5 forecasts (introduced in section 3 and defined as dH/dt). Shown are (a),(d) total ice thickness change and its decomposition into (b),(e) thermodynamic and (c),(f) advective contributions. Overlaid vectors in (c) and (f) indicate the climatology and trend, respectively, of ice velocity. Conversion factor from cm month−1 to W m−2 is ∼1.18.
Citation: Journal of Climate 35, 9; 10.1175/JCLI-D-21-0463.1
At the same time, Figs. 5d–f provide compelling evidence that many regional features of the Arctic winter sea ice changes in 1993–2020 were shaped by changes in sea ice dynamics rather than thermodynamic ice growth/melt (cf. Figs. 5f and 5d). It is corroborated by reasonably high [R = 0.51 (0.46, 0.57)] pattern correlation between spatial distributions of trends of sea ice thickness and its advective constituent (pattern correlation is estimated using all available spatially distributed pairs of values of two parameters; values in the parentheses represent the 95% confidence interval based on a bootstrapping method). Particularly, an accelerated Beaufort Gyre resulted in weaker sea ice convergence along the Canadian Arctic Archipelago and stronger convergence in the Amerasian Basin (e.g., Proshutinsky et al. 2009). One of the consequences of the accelerated ice drift along the southern flank of the Beaufort Gyre was westward advection of thick ice from the area north of the Canadian Arctic Archipelago. As we will see in the next section, these regionally different trends have strong implications for the predictability of seasonal evolution of Arctic sea ice. Finally, we note that thermodynamic and dynamic contributions to the sea ice thickness trend are not independent [cf. Figs. 5e and 5f; pattern correlation R = −0.86 (−0.88, −0.83)], which makes identification of drivers of Arctic sea ice changes difficult.
In the following section we evaluate pan-Arctic and regional effects of change of sea ice dynamics and ice growth/melt on seasonal predictability of the Arctic sea ice.
5. Predictability of seasonal sea ice thickness evolution
a. Dynamic and thermodynamic factors controlling predictability of Arctic sea ice
Figure 6 shows 1993–2020 averages in the predictable component (PC; see section 3) of the total sea ice thickness change, and contributions from thermodynamic and dynamic components, in forecasts initialized on 1 February. Analysis is carried out for winter (February–April) when sea ice experiences direct impact of heat from the ocean interior and no atmospheric shortwave radiative forcing, and for summer (May–September, but still based on the same forecasts started in February). Winter preconditioning is argued as an important factor for predictability of summer sea ice state (e.g., Day et al. 2014; Chevallier and Salas-Melia 2012). For reference, Fig. S1 in the online supplemental material shows a decomposition of PC into ensemble mean and total spread as well as its interdecadal changes. Note that PC of the total sea ice thickness is not the sum of the PCs of the thermodynamic and dynamic components as each PC represents a normalized quantity, and the separation provides a qualitative description only.
Maps of 1993–2020 mean (a)–(c) February–April and (d)–(f) May–September predictable components (PCs) of (a),(d) total ice thickness change PCTot and its decomposition into (b),(e) thermodynamic PCTh and (c),(f) advective PCAd contributions. The start date of the forecasts is 1 Feb.
Citation: Journal of Climate 35, 9; 10.1175/JCLI-D-21-0463.1
At regional scales, the 1993–2020 PC of total ice thickness change is dominated by the PC of ice advection in the central Arctic (Figs. 6a,c) as evidenced by their pattern correlation R = 0.95 (0.94, 0.96) in the area north of 60°N, whereas contribution of thermodynamic ice production increases southward and becomes notable in the marginal Arctic and sub-Arctic seas (Fig. 6b). It is important to note that the PC used in this study is a model-based metric that does not use observations. Nevertheless, when we define PC in observations as a correlation between ensemble mean forecasted and observed seasonal change of the total sea ice thickness divided by model spread (see also Eade et al. 2014) we find qualitatively similar results (not shown). Figure S1b shows that the Barents and Kara Seas seem to be the only places where ice production ensemble-mean variability σensmean is high away from the ice edge. This confirms the important role of ocean heat flux in shaping regional predictability of sea ice. Moreover, Figs. S1g–l confirm the finding of Tietsche et al. (2014) that advective contributions to thickness variability are amplified along the coasts (see their Fig. 3). The fact that this is much more pronounced for σtotal (Fig. S1i) than σensmean (Fig. S1c) suggests that this is unpredictable variability, presumably wind-driven.
Considering summer (May–September) PCs, we find that both dynamic (PCAd) and thermodynamic (PCTh) components of PC, as well as the total PC (PCTot), decay with progression from winter to summer (cf. Figs. 6a–c and 6d–f). While decay of PC at long lead times is expected, the lower PCs in summer do not stem from the long lead times. They are equally low for forecasts initialized in May (Fig. S2). This is aligned with the previous findings of lower skill in seasonal sea ice volume forecasts in summer (e.g., Day et al. 2014) due to the sea ice spring predictability barrier (Bushuk et al. 2020).
Unsurprisingly, the predictability of seasonal (growth-to-melt) evolution of pan-Arctic sea ice is dominated by thermodynamics as evidenced by close resemblance between total PCTot and its thermodynamic component PCTh (Fig. 7a). Both pan-Arctic PCTh and PCTot rapidly go down from PC ∼ 0.8 in May to PC ∼ 0.4 in July. This winter-to-summer pattern is consistent with seasonal evolution toward thinner, more mobile, and less predictable sea ice (Holland et al. 2008; Goosse et al. 2009, Blanchard-Wrigglesworth et al. 2011b). The dominance of PCTh over PCAd in shaping PCTot is expected since, at these spatial scales, the advection mostly redistributes sea ice within the domain without adding or removing ice. As we see next, this is not true at regional scales where advection becomes an important contributor to sea ice seasonal evolution and, thus, its predictability.
(a)–(c) Predictable component (PC), (d)–(f) variances of the ensemble means, and (g)–(i) total variances of sea ice thickness change (dH/dt)Tot and its decomposition into thermodynamic (dH/dt)Th and dynamic (dH/dt)Ad components averaged over (a),(d),(g) the entire Arctic, (b),(e),(h) the Eurasian Basin and (c),(f),(i) the Amerasian Basin. All values are based on reforecasts initialized on 1 Feb and averaged over 1993–2020. The values represent the PCs for accumulated changes since the beginning of the forecasts, i.e., September dH/dt represents the PC of thickening from February to September.
Citation: Journal of Climate 35, 9; 10.1175/JCLI-D-21-0463.1
Indeed, Figs. 7b and 7c show that both advection (convergence) and thermodynamics play an important role in shaping predictability of basin-scale seasonal sea ice changes. Predictability of the thermodynamic component PCTh is greater than predictability of the advective component PCAd in all months in both EB and AB (Figs. 7b,c). However, the variance (both total and ensemble mean) of advective tendencies is larger than the variance of thermodynamic tendencies (Figs. 7e,f,h,i), which gives the advective tendencies more weight in the PC of total tendencies. Covariances between advective and thermodynamic tendencies also play a role. In summary, while both advective and thermodynamic tendencies play an important role in shaping predictability of basin-scale seasonal sea ice changes, it seems that the advective tendencies are dominant.
Having a closer look at processes controlling sea ice predictability in the AB and EB, we note that observations and our modeling results demonstrate an acceleration of the Beaufort Gyre (particularly along its southern flank) in recent decades, which promotes more rapid advection of thick ice from the area north of the Canadian Archipelago. We hypothesize that this pattern dominates seasonal sea ice predictability in the AB in winter and summer (Figs. 6c,f). Processes associated with sea ice growth/melt due to increased influx of anomalously warm PW may be responsible for enhanced PCTh in this basin. At the same time, in the eastern EB, PCAd increases from February through April (Fig. 6c); however, Fig. 6f shows that starting from May, regional PCAd decreases due to loss of seasonal sea ice cover. Sea ice growth/melt drives contrasting regional changes of PCTh in the EB, with much stronger increase of PCTh in the Nansen Basin and weaker increase in the Amundsen Basin for February–April (Fig. 6b); in the latter basin, this weak increase is replaced by weakening of PCTh in May–September (Fig. 6e). This pattern is consistent with the pathway of the AW into the EB (Fig. 1).
b. Change of Arctic sea ice predictability in time
At pan-Arctic scales, PCTot of the total sea ice change shows an increase of predictability in winter [here defined as February–May (FMAM)] and decrease in summer [June–September (JJAS)] during the recent (2007–20) period compared with the earlier one (1993–2006) (Fig. 8a). This complex pattern is caused by a contrasting change in predictable components PCTh and PCAd. Since the mid-2000s, predictability expressed by advective component PCAd is reduced from March through April. It should be kept in mind that the advective component is computed indirectly (see section 3) and hence represents the residual of two large terms when considering pan-Arctic advection. This estimate thus may be contaminated by diagnostic inaccuracies and should be interpreted with care. Nevertheless, the generally reduced values of PCAd in the late period are plausible and in line with reduced Arctic sea ice export in recent years, which is also reflected by a substantial reduction by ensemble mean variance of ice advection (see Fig. 8d). The PCTh shows an increase throughout the growth-to-melt season (Fig. 8a). At the same time, both pan-Arctic ensemble mean and total variances of the thermodynamic component greatly exceed those of the advective component (Figs. 8d,g), suggesting that temporal change of PCTot is dominated by changes of PCTh.
(a)–(c) Predictable component (PC), (d)–(f) variances of the ensemble means, and (g)–(i) total variances of sea ice thickness change (dH/dt)Tot and its decomposition into thermodynamic (dH/dt)Th and dynamic (dH/dt)Ad components averaged over (a),(d),(g) the entire Arctic, (b),(e),(h) the Eurasian Basin, and (c),(f),(i) the Amerasian Basin. All values are based on reforecasts initialized on 1 Feb and averaged over 1993–2006 (early period; solid lines) and 2007–20 (late period; broken lines). The values represent the PCs for accumulated changes since the beginning of the forecasts, i.e., September dH/dt represents the PC of thickening from February to September.
Citation: Journal of Climate 35, 9; 10.1175/JCLI-D-21-0463.1
At regional scales, the change in time of the seasonal pattern of the AB PCTot bears similarities to the one of the pan-Arctic PCTot (Figs. 8a,c). PCTot in the AB is increased in February–March and again in September and decreased in summer months in the recent period, which is consistent with regional changes of PCAd and is possibly related to speedup of the Beaufort Gyre. The reduction of PCAd and consistent changes of PCTot in the AB at lead months from July through August in 2007–20 coincide with a strong decrease of ensemble mean variance of the advective contribution in summer (Fig. 8f). The latter may be related to the generally thinner ice in the recent period, which leads to a reduction of ensemble mean variance of advection. At the same time, the AB thermodynamic component of sea ice change shows a sizable increase of predictability in recent decades in all months of the ice growth-to-melt season (Fig. 8c). However, higher variance of the advective component suggests stronger influence of this component on the total predictability in the AB (Figs. 8e,f,h,i).
As in the AB, the EB PCTh shows an increase in recent decades at all lead times (which may be interpreted as a fingerprint of Atlantification) whereas predictability of advective component is reduced in all months over 2007–20 compared with 1993–2006. This, together with higher variances of the advective component (Figs. 8e,h), dominates the decline of predictability of the total sea ice change in the EB. Figure 8 also suggests that in both EB and AB and at pan-Arctic scale, variance of thermodynamic component increases only slightly but variance of advective component decreases strongly. That supports our finding that there is an increased predictability from thermodynamics, but it is overshadowed by strong changes in sea ice advection contribution to forecast uncertainty. Finally, we note that negative covariance between effects of advection and thermodynamics increases with lead time for ensemble means, but it stays relatively flat for total variance, suggesting an increasing negative contribution to total PC from the interaction of advection and thermodynamics.
6. Discussion and concluding remarks
In this paper, a combination of state-of-the-art eddy-permitting ocean-sea ice ensemble reanalysis ORAS5 and dynamic seasonal forecasting system SEAS5 is used to study effects of sea ice dynamics and thermodynamics on seasonal (growth-to-melt) Arctic sea ice predictability. The reanalysis product used to initialize the forecasts was extensively tested and demonstrated reasonable skill in representing major observed regional oceanic and sea ice changes. Particularly, this product was skillful in reproducing the observed intensification of ice drift associated with the Beaufort Gyre in the 1990s–2010s. The reanalysis showed the observed salinification of the EB halocline, with attendant reductions in stratification, and amplified upward heat fluxes. Also in accordance with observations (e.g., Proshutinsky et al. 2019; Polyakov et al. 2020b), the reanalysis showed opposing changes in the AB halocline where freshening increases stratification and limits communication between the SML and sea ice on one side and the ocean interior on another side. These reanalysis capabilities are critical considering the problems which current generation of the general circulation models have with simulating the evolution of Arctic halocline (e.g., Khosravi et al. 2022). As a word of caution, we add here that the reanalysis underestimated significantly the rates of salinification in the EB halocline and freshening in the AB halocline compared to available observations; that may lead to underestimating of the effects of thermodynamics on Arctic sea ice predictability in this study.
We demonstrate that thermodynamics (growth/melt) is dominant in defining the mean state of sea ice thickness, both at pan-Arctic and regional scales, as well as the seasonal (growth-to-melt season) predictability of the mean pan-Arctic sea ice in 1993–2020. We reiterate here, however, that thermodynamic and dynamic contributions to the sea ice thickness changes are not independent, which makes identification of drivers of Arctic sea ice changes difficult. These interactions between the thermodynamics and advection components require special attention and further studies. It may be plausible that a feedback mechanism in which winter convection in the EB, ice export, and regrowth in winter close to continental margins may play a role. Polyakov et al. (2020a) also suggested that a stronger dynamic and thermodynamic coupling between atmosphere, ice, and ocean in the eastern Arctic in the recent decade resulted in developing a feedback mechanism. However, feedbacks between the two components have likely a small impact on sea ice predictability, compared to individual contributions. The complex role of dynamic and thermodynamic forcings in shaping regional sea ice concentrations north of Svalbard and east of Greenland was also discussed by Lundesgaard et al. (2021) and Våge at al. (2018), respectively.
Analysis of forecasts initialized in February show that pan-Arctic predictability of both total sea ice change and its thermodynamic component decreases steadily from winter to summer; and the relatively low predictability of summer sea ice is also evident in forecasts initialized in May. This is probably related to the spring predictability barrier in which negative feedbacks from sea ice growth play an important role (Tietsche et al. 2011; Bushuk et al. 2020) as well as to increased influence of atmospheric fluxes in summer as indicated by increase of total variance in summer (Fig. 4a; see also Tietsche et al. 2015). Evolution of predictability of advective component at pan-Arctic scales with lead time is not monotonic, reaching a maximum in June and a minimum in August, which is consistent with seasonal evolution toward thicker, less mobile, and more predictable sea ice in winter (Holland et al. 2011). On the other hand, at regional scales predictability of sea ice is dominated by sea ice dynamics, while the contribution from sea ice growth/melt remains perceptible.
Change of predictability of sea ice is a fingerprint of climate change. The existing analyses are not conclusive, however, as they are strongly dependent on model setup, initial conditions, and region, and whether sea ice area, thickness, or volume are examined [see reviews, e.g., Guemas et al. (2014), for discussion therein]. Our results demonstrate competing influences of sea ice dynamics and thermodynamics on temporal (1993–2020) change of predictability with increasing predictability caused by thermodynamic growth, attributed to increased late-winter to spring ocean heat flux, and decreasing predictability due to advection. In that, the role of advection increases as the region of interest becomes smaller. Indeed, we show that advection of sea ice has no impact on seasonal sea ice predictability at pan-Arctic scale (Figs. 7a and 8a), with considerable influence on basin scales (EB/AB; Figs. 7b,c and 8b,c), dominating predictability at smaller regional scales (Fig. 6). To verify that our results are not model dependent, the approach of this study should be repeated with other forecast systems.
Previous studies evaluated the role of various factors in restraining Arctic sea ice predictability (see section 1 and the references therein). In this study, for the first time, the role of thermodynamic and dynamic contributions to change of Arctic sea ice predictability was evaluated and the role of each factor at pan-Arctic and regional scales is accentuated. Sea ice drift is often overlooked when evaluating sea ice forecasts, but given our results it probably needs much more attention in developing reliable regional sea ice predictions. Furthermore, with a projected increase of freshening of the upper Arctic Ocean in the twenty-first century (Khosravi et al. 2022), the enhanced oceanic stratification would lead to reduced ocean heat fluxed from the ocean interior to the bottom of sea ice, thus further strengthening the role of advection in shaping Arctic sea ice and reducing summer sea ice predictability. Finally, we note that this study delivers critical information about sea ice predictability in the “new Arctic” conditions, increasing awareness regarding sea ice state and implementation of sea ice forecasts for the needs of shipping, high-latitude tourism, fishing, industry, etc., which all depend on accurate seasonal sea ice forecasts.
Acknowledgments.
IP contribution was supported by National Science Foundation Grants 1708424, 1708427, and 1724523. MM’s contribution was funded by EU H2020 Grant agreements 821984 (KEPLER) and 862626 (EUROSEA) as well as Austrian Science Fund project P33177.
Data availability statement.
The data associated with this paper are available from the authors upon request.
REFERENCES
Athanase, M., C. Provost, M. D. Pérez-Hernández, N. Sennéchael, C. Bertosio, C. Artana, G. Garric, and J.-M. Lellouche, 2020: Atlantic water modification north of Svalbard in the Mercator physical system from 2007 to 2020. J. Geophys. Res. Oceans, 125, e2020JC016463, https://doi.org/10.1029/2020JC016463.
Babb, D. G., J. C. Landy, J. V. Lukovich, C. Haas, S. Hendricks, D. G. Barber, and R. J. Galley, 2020: The 2017 reversal of the Beaufort Gyre: Can dynamic thickening of a seasonal ice cover during a reversal limit summer ice melt in the Beaufort Sea? J. Geophys. Res. Oceans, 125, e2020JC016796, https://doi.org/10.1029/2020JC016796.
Ballinger, T. J., and Coauthors, 2021: Unusual west Arctic storm activity during winter 2020: Another “collapse” of the Beaufort high? Geophys. Res. Lett., 48, e2021GL092518, https://doi.org/10.1029/2021GL092518.
Blanchard-Wrigglesworth, E., K. C. Armour, C. M. Bitz, and E. DeWeaver, 2011a: Persistence and inherent predictability of Arctic sea ice in a GCM ensemble and observations. J. Climate, 24, 231–250, https://doi.org/10.1175/2010JCLI3775.1.
Blanchard-Wrigglesworth, E., C. M. Bitz, and M. M. Holland, 2011b: Influence of initial conditions and climate forcing on predicting Arctic sea ice. Geophys. Res. Lett., 38, L18503, https://doi.org/10.1029/2011GL048807.
Blanchard-Wrigglesworth, E., and Coauthors, 2017: Multi-model seasonal forecast of Arctic sea-ice: Forecast uncertainty at pan-Arctic and regional scales. Climate Dyn., 49, 1399–1410, https://doi.org/10.1007/s00382-016-3388-9.
Bourke, R. H., and R. P. Garrett, 1987: Sea ice thickness distribution in the Arctic Ocean. Cold Reg. Sci. Technol., 13, 259–280, https://doi.org/10.1016/0165-232X(87)90007-3.
Bushuk, M., M. Winton, D. B. Bonan, E. Blanchard-Wrigglesworth, and T. L. Delworth, 2020: A mechanism for the Arctic sea ice spring predictability barrier. Geophys. Res. Lett., 47, e2020GL088335, https://doi.org/10.1029/2020GL088335.
Carmack, E., and Coauthors, 2015: Toward quantifying the increasing role of oceanic heat in sea ice loss in the new Arctic. Bull. Amer. Meteor. Soc., 96, 2079–2105, https://doi.org/10.1175/BAMS-D-13-00177.1.
Chevallier, M., and D. Salas-Melia, 2012: The role of sea ice thickness distribution in the Arctic sea ice potential predictability: A diagnostic approach with a coupled GCM. J. Climate, 25, 3025–3038, https://doi.org/10.1175/JCLI-D-11-00209.1.
Danielson, S. L., and Coauthors, 2020: Manifestation and consequences of warming and altered heat fluxes over the Bering and Chukchi Sea continental shelves. Deep-Sea Res. II, 177, 104781, https://doi.org/10.1016/j.dsr2.2020.104781.
Day, J. J., E. Hawkins, and S. Tietsche, 2014: Will Arctic sea ice thickness initialization improve seasonal forecast skills? Geophys. Res. Lett., 41, 7566–7575, https://doi.org/10.1002/2014GL061694.
Dee, D. P., and Coauthors, 2011: The ERA‐Interim reanalysis: Configuration and performance of the data assimilation system. Quart. J. Roy. Meteor. Soc., 137, 553–597, https://doi.org/10.1002/qj.828.
Eade, R., D. Smith, A. Scaife, E. Wallace, N. Dunstone, L. Hermanson, and N. Robinson, 2014: Do seasonal‐to‐decadal climate predictions underestimate the predictability of the real world? Geophys. Res. Lett., 41, 5620–5628, https://doi.org/10.1002/2014GL061146.
Eicken, H., 2013: Arctic sea ice needs better forecasts. Nature, 497, 431–433, https://doi.org/10.1038/497431a.
Goosse, H., O. Arzel, C. M. Bitz, A. de Montety, and M. Vancoppenolle, 2009: Increased variability of the Arctic summer ice extent in a warmer climate. Geophys. Res. Lett., 36, L23702, https://doi.org/10.1029/2009GL040546.
Graversen, R. G., T. Mauritsen, S. Drijfhout, M. Tjernström, and S. Mårtensson, 2011: Warm winds from the Pacific caused extensive Arctic sea-ice melt in summer 2007. Climate Dyn., 36, 2103–2112, https://doi.org/10.1007/s00382-010-0809-z.
Guemas, V., and Coauthors, 2014: A review on Arctic sea-ice predictability and prediction on seasonal to decadal time-scales. Quart. J. Roy. Meteor. Soc., 142, 546–561, https://doi.org/10.1002/qj.2401.
Hamilton, L. C., and J. Stroeve, 2016: 400 predictions: The SEARCH Sea Ice Outlook 2008–2015. Polar Geogr., 39, 274–287, https://doi.org/10.1080/1088937X.2016.1234518.
Holland, M. M., C. M. Bitz, B. Tremblay, and D. A. Bailey, 2008: The role of natural versus forced change in future rapid summer Arctic ice loss. Arctic Sea Ice Decline: Observations, Projections, Mechanisms and Implications. Geophys. Monogr., Vol. 180, Amer. Geophys. Union, 133–150.
Holland, M. M., D. A. Bailey, and S. Vavru, 2011: Inherent sea ice predictability in the rapidly changing Arctic environment of the Community Climate System Model version 3. Climate Dyn., 36, 1239–1253, https://doi.org/10.1007/s00382-010-0792-4.
Holland, M. M., L. Landrum, D. Bailey, and S. Vavrus, 2019: Changing seasonal predictability of Arctic summer sea ice area in a warming climate. J. Climate, 32, 4963–4979, https://doi.org/10.1175/JCLI-D-19-0034.1.
Ivanov, V., A. Smirnov, V. Alexeev, N. V. Koldunov, I. Repina, and V. Semenov, 2018: Contribution of convection‐induced heat flux to winter ice decay in the western Nansen Basin. J. Geophys. Res. Oceans, 123, 6581–6597, https://doi.org/10.1029/2018JC013995.
Johnson, S. J., and Coauthors, 2019: SEAS5: The new ECMWF Seasonal Forecast System. Geosci. Model Dev., 12, 1087–1117, https://doi.org/10.5194/gmd-12-1087-2019.
Khosravi, N., Q. Wang, N. Koldunov, C. Hinrichs, T. Semmler, S. Danilov, and T. Jung, 2022: Arctic Ocean in CMIP6 models: Historical and projected temperature and salinity in the deep basins. Earth’s Future, https://www.essoar.org/doi/10.1002/essoar.10507417.1; in press.
Kwok, R., 2015: Sea ice convergence along the Arctic coasts of Greenland and the Canadian Arctic Archipelago: Variability and extremes (1992–2014). Geophys. Res. Lett., 42, 7598–7605, https://doi.org/10.1002/2015GL065462.
Kwok, R., 2018: Arctic sea ice thickness, volume, and multiyear ice coverage: Losses and coupled variability (1958–2018). Environ. Res. Lett., 13, 105005, https://doi.org/10.1088/1748-9326/aae3ec.
Kwok, R., and N. Untersteiner, 2011: The thinning of Arctic sea ice. Phys. Today, 64, 36–41, https://doi.org/10.1063/1.3580491.
Labe, Z., G. Magnusdottir, and H. Stern, 2018: Variability of Arctic sea ice thickness using PIOMAS and the CESM large ensemble. J. Climate, 31, 3233–3247, https://doi.org/10.1175/JCLI-D-17-0436.1.
Lindsay, R., and A. Schweiger, 2015: Arctic sea ice thickness loss determined using subsurface, aircraft, and satellite observations. Cryosphere, 9, 269–283, https://doi.org/10.5194/tc-9-269-2015.
Lindsay, R., J. Zhang, A. J. Schweiger, and M. A. Steele, 2008: Seasonal predictions of ice extent in the Arctic Ocean. J. Geophys. Res., 113, C02023, https://doi.org/10.1029/2007JC004259.
Lundesgaard, Ø., A. Sundfjord, and A. H. H. Renner, 2021: Drivers of interannual sea ice concentration variability in the Atlantic Water inflow region north of Svalbard. J. Geophys. Res. Oceans, 126, e2020JC016522, https://doi.org/10.1029/2020JC016522.
Mayer, M., S. Tietsche, L. Haimberger, T. Tsubouchi, J. Mayer, and H. Zuo, 2019: An improved estimate of the coupled Arctic energy budget. J. Climate, 32, 7915–7934, https://doi.org/10.1175/JCLI-D-19-0233.1.
McPhee, M. G., A. Proshutinsky, J. H. Morison, M. Steele, and M. B. Alkire, 2009: Rapid change in freshwater content of the Arctic Ocean. Geophys. Res. Lett., 36, L10602, https://doi.org/10.1029/2009GL037525.
Melia, N., K. Haines, and E. Hawkins, 2016: Sea ice decline and 21st century trans-Arctic shipping routes. Geophys. Res. Lett., 43, 9720–9728, https://doi.org/10.1002/2016GL069315.
Moore, G. W. K., A. Schweiger, J. Zhang, and M. Steele, 2018: Collapse of the 2017 winter Beaufort high: A response to thinning sea ice? Geophys. Res. Lett., 45, 2860–2869, https://doi.org/10.1002/2017GL076446.
Morison, J., R. Kwok, C. Peralta-Ferriz, M. Alkire, I. Rigor, R. Andersen, and M. Steele, 2012: Changing Arctic Ocean freshwater pathways. Nature, 481, 66–70, https://doi.org/10.1038/nature10705.
Notz, D., and J. Stroeve, 2016: Observed Arctic sea-ice loss directly follows anthropogenic CO2 emission. Science, 354, 747–750, https://doi.org/10.1126/science.aag2345.
Pérez‐Hernández, M. D., and Coauthors, 2019: Structure, transport, and seasonality of the Atlantic Water boundary current north of Svalbard: Results from a yearlong mooring array. J. Geophys. Res. Oceans, 124, 1679–1698, https://doi.org/10.1029/2018JC014759.
Petty, A. A., D. Schröder, J. C. Stroeve, T. Markus, J. Miller, N. T. Kurtz, D. L. Feltham, and D. Flocco, 2017: Skillful spring forecasts of September Arctic sea ice extent using passive microwave sea ice observations. Earth’s Future, 5, 254–263, https://doi.org/10.1002/2016EF000495.
Polyakov, I. V., D. Walsh, I. Dmitrenko, R. Colony, and L. Timokhov, 2003: Arctic Ocean variability derived from historical observations. Geophys. Res. Lett., 30, 1298, https://doi.org/10.1029/2002GL016441.
Polyakov, I. V., and Coauthors, 2017: Greater role for Atlantic inflows on sea-ice loss in the Eurasian Basin of the Arctic Ocean. Science, 356, 285–291, https://doi.org/10.1126/science.aai8204.
Polyakov, I. V., A. V. Pnyushkov, and E. C. Carmack, 2018: Stability of the Arctic halocline: A new indicator of Arctic climate change. Environ. Res. Lett., 13, 125008, https://doi.org/10.1088/1748-9326/aaec1e.
Polyakov, I. V., and Coauthors, 2020a: Weakening of cold halocline layer exposes sea ice to oceanic heat in the eastern Arctic Ocean. J. Climate, 33, 8107–8123, https://doi.org/10.1175/JCLI-D-19-0976.1.
Polyakov, I. V., and Coauthors, 2020b: Borealization of the Arctic Ocean in response to anomalous advection from sub-Arctic seas. Front. Mar. Sci., 7, 491, https://doi.org/10.3389/fmars.2020.00491.
Proshutinsky, A., and Coauthors, 2009: Beaufort Gyre fresh water reservoir: State and variability from observations. J. Geophys. Res., 114, C00A10, https://doi.org/10.1029/2008JC005104.
Proshutinsky, A., and Coauthors, 2019: Analysis of the Beaufort Gyre freshwater content in 2003–2018. J. Geophys. Res. Oceans, 124, 9658–9689, https://doi.org/10.1029/2019JC015281.
Rampal, P., J. Weiss, and D. Marsan, 2009: Positive trend in the mean speed and deformation rate of Arctic sea ice, 1979–2007. J. Geophys. Res., 114, C05013, https://doi.org/10.1029/2008JC005066.
Rampal, P., J. Weiss, C. Dubois, and J. M. Campin, 2011: IPCC climate models do not capture Arctic sea ice drift acceleration: Consequences in terms of projected sea ice thinning and decline. J. Geophys. Res., 116, C00D07, https://doi.org/10.1029/2011JC007110.
Rigor, I., and J. M. Wallace, 2004: Variations in the age of Arctic sea-ice and summer sea-ice extent. Geophys. Res. Lett., 31, L09401, https://doi.org/10.1029/2004GL019492.
Schröder, D., D. Feltham, D. Flocco, and M. Tsamados, 2014: September Arctic sea-ice minimum predicted by spring melt-pond fraction. Nat. Climate Change, 4, 353–357, https://doi.org/10.1038/nclimate2203.
Serreze, M. C., and A. P. Barrett, 2008: The summer cyclone maximum over the central Arctic Ocean. J. Climate, 21, 1048–1065, https://doi.org/10.1175/2007JCLI1810.1.
Serreze, M. C., A. D. Crawford, J. C. Stroeve, A. P. Barrett, and R. A. Woodgate, 2016a: Variability, trends, and predictability of seasonal sea ice retreat and advance in the Chukchi Sea. J. Geophys. Res. Oceans, 121, 7308–7325, https://doi.org/10.1002/2016JC011977.
Serreze, M. C., J. C. Stroeve, A. P. Barrett, and L. N. Boisvert, 2016b: Summer atmospheric circulation anomalies over the Arctic Ocean and their influences on September sea ice extent: A cautionary tale. J. Geophys. Res. Atmos., 121, https://doi.org/10.1002/2016JD025161.
Shimada, K., T. Kamoshida, M. Itoh, S. Nishino, E. Carmack, F. McLaughlin, S. Zimmermann, and A. Proshutinsky, 2006: Pacific Ocean inflow: Influence on catastrophic reduction of sea ice cover in the Arctic Ocean. Geophys. Res. Lett., 33, L08605, https://doi.org/10.1029/2005GL025624.
Sigmond, M., J. C. Fyfe, and N. C. Swart, 2018: Ice-free Arctic projections under the Paris Agreement. Nat. Climate Change, 8, 404–408, https://doi.org/10.1038/s41558-018-0124-y.
Simmonds, I., and K. Keay, 2009: Extraordinary September Arctic sea ice reductions and their relationships with storm behavior over 1979–2008. Geophys. Res. Lett., 36, L19715, https://doi.org/10.1029/2009GL039810.
Sorteberg, A., and J. E. Walsh, 2008: Seasonal cyclone variability at 70°N and its impact on moisture transport into the Arctic. Tellus, 60A, 570–586, https://doi.org/10.1111/j.1600-0870.2008.00314.x.
Spreen, G., R. Kwok, and D. Menemenlis, 2011: Trends in Arctic sea ice drift and role of wind forcing: 1992–2009. Geophys. Res. Lett., 38, L19501, https://doi.org/10.1029/2011GL048970.
Stroeve, J., and D. Notz, 2018: Changing state of Arctic sea ice across all seasons. Environ. Res. Lett., 13, 103001, https://doi.org/10.1088/1748-9326/aade56.
Stroeve, J., A. D. Craford, and S. Stammerjohn, 2016: Using timing of ice retreat to predict timing of fall freeze‐up in the Arctic. Geophys. Res. Lett., 43, 6332–6340, https://doi.org/10.1002/2016GL069314.
Thompson, D. W. J., and J. M. Wallace, 1998: The Arctic Oscillation signature in the wintertime geopotential height and temperature fields. Geophys. Res. Lett., 25, 1297–1300, https://doi.org/10.1029/98GL00950.
Tietsche, S., D. Notz, J. H. Jungclaus, and J. Marotzke, 2011: Recovery mechanisms of Arctic summer sea ice. Geophys. Res. Lett., 38, L02707, https://doi.org/10.1029/2010GL045698.
Tietsche, S., D. Notz, J. H. Jungclaus, and J. Marotzke, 2013: Predictability of large interannual Arctic sea-ice anomalies. Climate Dyn., 41, 2511–2526, https://doi.org/10.1007/s00382-013-1698-8.
Tietsche, S., and Coauthors, 2014: Seasonal to interannual Arctic sea-ice predictability in current global climate models. Geophys. Res. Lett., 41, 1035–1043, https://doi.org/10.1002/2013GL058755.
Tietsche, S., E. Hawkins, and J. J. Day, 2015: Atmospheric and oceanic contributions to irreducible forecast uncertainty of Arctic surface climate. J. Climate, 29, 331–346, https://doi.org/10.1175/JCLI-D-15-0421.1.
Tschudi, M. A., and Coauthors, 2020: An enhancement to sea ice motion and age products at the National Snow and Ice Data Center (NSIDC). Cryosphere, 14, 1519–1536, https://doi.org/10.5194/tc-14-1519-2020.
Uotila, P., and Coauthors, 2019: An assessment of ten ocean reanalyses in the polar regions. Climate Dyn., 52, 1613–1650, https://doi.org/10.1007/s00382-018-4242-z.
Våge, K., L. Papritz, L. Håvik, M. A. Spall, and G. W. K. Moore, 2018: Ocean convection linked to the recent ice edge retreat along east Greenland. Nat. Commun., 9, 1287, https://doi.org/10.1038/s41467-018-03468-6.
Wang, M., and J. E. Overland, 2012: A sea ice free summer Arctic within 30 years: An update from CMIP5 models. Geophys. Res. Lett., 39, L18501, https://doi.org/10.1029/2012GL052868.
Woodgate, R. A., 2018: Increases in the Pacific inflow to the Arctic from 1990 to 2015, and insights into seasonal trends and driving mechanisms from year-round Bering Strait mooring data. Prog. Oceanogr., 160, 124–154, https://doi.org/10.1016/j.pocean.2017.12.007.
Wu, B., J. Wang, and J. E. Walsh, 2006: Dipole anomaly in the winter Arctic atmosphere and its association with sea ice motion. J. Climate, 19, 210–225, https://doi.org/10.1175/JCLI3619.1.
Yamamoto-Kawai, M., F. A. McLaughlin, E. C. Carmack, S. Nishino, K. Shimada, and N. Kurita, 2009: Surface freshening of the Canada Basin, 2003–2007: River runoff versus sea ice meltwater. J. Geophys. Res., 114, C00A05, https://doi.org/10.1029/2008JC005000.
Zhang, X., J. E. Walsh, J. Zhang, U. S. Bhatt, and M. Ikeda, 2004: Climatology and interannual variability of Arctic cyclone activity, 1948–2002. J. Climate, 17, 2300–2317, https://doi.org/10.1175/1520-0442(2004)017<2300:CAIVOA>2.0.CO;2.
Zhang, X., A. Sorteberg, J. Zhang, R. Gerdes, and J. C. Comiso, 2008: Recent radical shifts in atmospheric circulations and rapid changes in Arctic climate system. Geophys. Res. Lett., 35, L22701, https://doi.org/10.1029/2008GL035607.
Zuo, H., M. A. Balmaseda, S. Tietsche, K. Mogensen, and M. Mayer, 2019: The ECMWF operational ensemble reanalysis–analysis system for ocean and sea ice: A description of the system and assessment. Ocean Sci., 15, 779–808, https://doi.org/10.5194/os-15-779-2019.