Contribution of the deepened Amundsen sea low to the record low Antarctic sea ice extent in February 2022

Sea ice & SST: Monthly mean sea ice concentration (SIC) for the period 1979–2022 is obtained from the National Snow and Ice Data Center (NSIDC). This dataset is derived from Special Sensor Microwave Imager/Sounder (SSMIS) on board the Defense Meteorological Satellite Program (DMSP) satellites using the NASA team algorithm [20]. The daily SIC is obtained from the NASA team near-real time product. The data are provided on the polar stereographic projection with a spatial resolution of 25 km. Monthly sea ice drift is from the similar 25 km Polar Pathfinder Equal-Area Scalable Earth (EASE)-Grid, which is based on optimal interpolation scheme to combine satellite imagery, reanalysis winds and buoy position data [21]. Monthly sea surface temperature (SST) for the period 1982–2022 is taken from the National Oceanic and Atmospheric Administration (NOAA) with a grid resolution of 1° × 1° [22].

CERES: Monthly surface radiative fluxes are obtained from the Clouds and the Earth's Radiant Energy System (CERES) from March 2000 to February 2022. The surface radiative fluxes are extracted from level-2B FLXHR product [23], which are provided at the 1° by 1° grid.

ERA5 & ORAS5: The atmospheric circulation fields used in this study are from the ERA5 reanalysis for the period 1979–2022, including the monthly mean sea level pressure (SLP), 10 meter zonal and meridional winds, geopotential height (GPH) at 500 hPa and 50 hPa, shortwave and longwave radiative fluxes at the surface [24]. The SLP over the period 1950–1978 is retrieved from the ERA5 extended reanalysis. The ocean temperature data for the common period are from the ocean and sea-ice reanalyses (ORAs, or ocean syntheses) of version 5 (ORAS5), which is a global eddy-permitting ocean and sea-ice ensemble reanalysis [25]. We calculated the mean temperature above 50 m depth to represent the upper ocean temperature variability.

CMIP6 historical run: A key element of CMIP Phase 6 (CMIP6) [26] is the simulation of climate change observed during the so-called historical period from 1850 to 2014. Simulations of 30 models available in CMIP6 historical run and the model outputs of SIE, SIC and atmospheric variables are used. The salient information of these models is listed in the supplementary material (table S1).

CESM2-WACCM: CESM2 represents the Community Earth System Model version 2, while the WACCM is the atmospheric component model and represents the Whole Atmosphere Community Climate Model version 6, which considers interactive stratospheric chemistry and gravity wave parameterizations in the upper atmosphere [27]. Compared to its previous version CESM1, Antarctic sea ice area in CESM2 is in better agreement with the satellite era observations, particularly in terms of maximum and minimum area [28]. Overall, the CESM2-WACCM simulates a very consistent representation of the seasonal minimum and maximum of the Antarctic SIE compared to the observations, which are still not well represented in many Earth system models (figure S3) [29].

Sea-ice advection: Similar to previous studies [30, 31], the wind-driven sea-ice advection is calculated as follows,

$$\begin{equation*}{\text{seaice advection = }}C \cdot V,\end{equation*}$$

where the variables $C$ and $V$ denote the SIC and the meridional component of sea ice motion, respectively. The original curvilinear grids are firstly interpolated to rectangular grids of 0.5° × 0.5° by distance weighted interpolation. Then, the sea-ice area flux across the longitude band at 70° S over the eastern Ross Sea (180° E–230° E) is estimated as follows,

$$\begin{equation*}{\text{area flux = }}\sum\limits_{i = 1}^n {{c_i}} {v_i}\Delta L,\end{equation*}$$

where ${c_i}$ and ${v_i}$ denotes the SIC and sea ice motion at each grid point, respectively; n denotes the number of intervals along the longitude band at 70° S and $\Delta L$ denotes the length of each interval.

Climate indices: Sea ice extent (SIE) is defined as the total area covered by at least 15% sea ice concentration (SIC). For the satellite record, Antarctic SIE is obtained from the NSIDC; for the CMIP6 historical simulations, the SIEs are manually calculated or retrieved from the webpage (https://esgf-node.llnl.gov/search/cmip6/). Similar to [32], ASL metrics including ASL position in latitude and longitude, ASL strength, ASL depth and relative strength are applied. Here, the ASL strength is defined as the average of SLP over the domain: 170° E–298° E and 80° S–60° S, and ASL depth is defined as the minimum SLP over the same domain. The data are obtained from https://github.com/scotthosking/amundsen-sea-low-index. We define the near-surface southerly winds over the eastern Ross Sea as the area-averaged meridional wind at 850 hPa (V850) over the domain (180° E–230° E and 80° S–60° S). The Niño 3.4 index is the area-averaged SST anomaly from 5° S–5° N in latitude and 170° E–120° W in longitude. The strength of the stratospheric polar vortex in the SH is computed as the polar-cap-averaged (poleward of 70° S) geopotential height at 50 hPa (PCH50) [33, 34]. Surface radiative fluxes, including net radiative flux (NetRad), net shortwave and longwave radiative flux (NetSW and NetLW), downward and upward shortwave radiation (DSW and USW), downward and upward longwave radiation (DLW and ULW), are applied and defined positive downward.

Composite analysis: Based on the CESM2-WACCM, the response of Antarctic sea ice and circulations to the ASL is estimated by the composite difference according to the 10th/90th percentile of the ASL strength.

To start with, for the 43 years from 1979 to 2022, the observed correlation coefficient between the SON ASL strength and Feb Antarctic SIE is about 0.53, regression coefficient or slope is 0.072 with uncertainty range of ±0.037 (figure 1(f)), estimated by 95% confidence intervals (grey lines). According to the linear regression model, corresponding to –8 hPa decrease in ASL strength, there is about 0.58 (±0.30) 10⁶ km² decrease in Feb Antarctic SIE, and thus contributing to about 60% of the Antarctic SIE anomaly in Feb 2022. While we examine the range of ASL metrics, we find that the ASL depth displays the strongest correlation (r = 0.63), and ASL latitude and longitude do not show a significant correlation with February SIE (not shown). It is also important to note that the correlations between SON ASL strength and monthly Antarctic SIE (September to the following August) are largest in February and statistically significant only in February and March (figure S4). This is when SIE generally reaches the annual minimum and regional sea ice variations in the Ross-Amundsen-Weddell Seas dominate the overall Antarctic SIE changes (figure 1(a)). Regionally, SON ASL strength is also highly correlated with the Weddell Sea SIE, but to a lesser extent with the Ross Sea SIE (figures 2(b) and (c)). Secondly, the lagged relationships are also examined in CMIP6 historical simulations. The correlations between SON ASL and Feb Antarctic, Ross Sea and Weddell Sea SIE over 1979–2014 (36 years) based on 30 CMIP6 models are shown in figure 2 (blue bars). Overall, compared to the other models, the CESM2 family represents well the relationship between the ASL and regional and total Antarctic SIE, which is consistent with the observations. Of the 30 models, 25 have a lower correlation than the observations, and 19 have a confidence interval that does not even overlap with that of the observations.

Zoom In Zoom Out Reset image size

The observed and well-simulated significant correlations between the SON ASL and the following Feb SIE indicate that the local circulation changes associated with a deepened ASL may have an impact on the seasonal transitions of Antarctic sea ice from spring to summer. To explore the underlying mechanism of this lagged impact, both observational data and CESM2-WACCM simulations are further examined.

In comparison with the deepest ASL in 2021, the 2002 ASL anomaly peaks with positive 8 hPa SLP anomalies, i.e. the shallowest ASL (figure 1(e)). Consistently, it is related to anticyclonic wind anomalies centred in the Amundsen Sea (not shown). To better depict the role of extreme ASL phase in shaping the regional sea ice, the evolution of SIC anomalies from SON to February (in the following year) in 2021/2022 and 2002/2003 are shown in figure 3 (upper panel). Over the Bellingshausen and northern Weddell Seas, there is statistically significant SIC decrease by 30% (>1 standard deviation of the interannual variability) in SON 2021 (figure 3(a)). Following the seasonal retreat of Antarctic sea ice from Dec to Feb, the sea ice margins continue to shrink southwards (figures 3(b)–(d)), while the sea ice over the interior (southward of 70° S, close to the Antarctic continent) starts to grow. Over the northern Ross and Amundsen Seas, the SON sea ice increases slightly at the edge and decreases over the interior. In the following Dec, Jan and Feb (DJF), the sea ice retreat in the interior of the Ross Sea becomes notable and significant (>40% SIC). As a result, the drastic sea ice loss at the edge of northern Weddell Sea and the inner eastern Ross Sea leads directly to the record low Antarctic SIE (figure 1(d)). In contrast to the deepest ASL in SON 2021, the ASL is shallowest in SON 2002 (figure 1(e)). Accordingly, there is a significant sea ice gain at the edge of the northern Weddell Sea and in the interior of the Ross Sea in DJF 2003.

Zoom In Zoom Out Reset image size

These relationships are further examined in CESM2-WACCM historical simulations (165 years). Based on the 10th/90th percentile of the ASL in CESM2-WACCM, we calculated the composite SIC anomalies from SON to February (in the following year). The 10th percentile corresponds to a deep ASL whereas the 90th percentile corresponds to a shallow ASL. The corresponding sea ice anomalies from spring to the following Feb are strongly consistent with observations (figure 3 lower panel vs. Figure 3 upper panel). For example, in response to the deeper ASLs of the simulations, the Feb sea ice experiences a substantial decrease over the northern Weddell and eastern Ross Seas and a notable sea ice increase in the interior of the Bellingshausen and Weddell Seas. These results indicate that the two extreme phases of the ASL can have opposing effects on sea ice evolution in the interior and exterior of the Weddell-Ross Seas.

We hypothesize that the lagged Antarctic SIC responses to the preceding ASL variability are related to wind-driven sea ice export over the Ross Sea. Firstly, corresponding to strong (deep) ASLs, there is a significant increase in southerly winds over the Ross Sea with correlations about –0.67 for observations and –0.51 for CESM2-WACCM, respectively (figures 4(a) and (b)). Associated with the deep ASL and the enhanced southerly winds, there is significant northward sea-ice advection over the eastern Ross Sea in spring (figures 4(c) and (d)). The correlation between the ASL and sea-ice area flux across the latitude band at 70° S is –0.63 in observations (and –0.57 in CESM2-WACCM). The deepest ASL in 2021 is associated with substantially enhanced meridional sea ice drift over the Ross Sea with values about 5.7 km day⁻¹ and sea-ice area flux across 70° S with values about 17.6 × 10³ km² in October-November; in contrast, the weakest ASL in 2002 is accompanied by meridional sea ice drift of about 0.4 km d⁻¹ over the Ross Sea and sea-ice area flux of about 0.6 × 10³ km² in October-November.

Zoom In Zoom Out Reset image size

Secondly, as the sea ice is expelled away from the coastline in spring, it is associated with an increase in open water areas near the coastline, i.e. the expansion of coastal polynyas. The 'ice factory' of coastal polynyas loses its effectiveness, when the temperature rises above freezing point from spring to summer. Then, less upward shortwave would be reflected to space in summer (figures 5(a) and (i)), resulting in an increase of solar radiation absorption by the open water areas (figures 5(b) and (j)). Consequently, it leads to a substantial warming in the upper ocean (figure 5(c)), which in turn accelerates the sea ice melting. This positive ice-ocean albedo feedback sustains the surface ice melting in subsequent months. In the month-latitude plane, associated with a deep ASL, a meridional dipole structure is observed over the Ross Sea with sea ice gain in the exterior from Oct to Jan but sea ice loss in the interior from Nov to March. Analogously, the model (CESM2-WACCM) reproduced the whole process chain as seen in observations (figure 5, upper panel vs. lower panel) as well as the sea ice dipole distribution over the Ross Sea. In response to the strongest (weakest) ASL in 2021 (2002), there is about 20–30 W m⁻² gain (loss) of net surface radiation (40% more than the climatology), and about 0.5 K upper ocean warming (cooling) over the eastern Ross Sea (figure S5).

Zoom In Zoom Out Reset image size

As for the northern Weddell Sea, the sea ice evolution over the Weddell Sea is different, with a continuous sea ice decrease at the edge from spring to summer and a slight increase in the interior Weddell Sea during DJF (figures S6(d) and (h)). The ice-ocean albedo feedback also sustains the loss of sea ice from spring to summer, associated with a stronger ASL (figure S6). However, this ice-ocean albedo feedback is initiated by warmer and moister air and northerly winds, and the accompanying increase of downward longwave radiation over the Weddell Sea in spring (figure S7), which has been well documented in previous studies [35, 36].

Previous studies have suggested that tropical SST variability is a major contributor to the ASL variability by generating Rossby wave trains that propagate to SH high latitudes [7, 37]. The spatial pattern of SLP anomalies in 2021 resembles the Pacific–South American (PSA) pattern. The equatorial SST over the Niño 3.4 region in SON 2021 shows significant cold anomalies (figure 6(a)), indicating La Niña conditions. The eastern and central Pacific cooling would induce a horizontal convergence in the upper troposphere with descending air, exciting a Rossby wave train that travels to the extra-tropics and deepens the ASL [38]. In comparison, the equatorial SST over the Niño 3.4 region in SON 2002 shows significant warm anomalies (figure 6(b)), indicating El Niño conditions, which leads to a shallower ASL. In 2002, the anomaly pattern is slightly different from that of 2021, which is related to Central Pacific El Niño conditions [39]. The eastward wave propagation from mid-to-high latitudes excited by the tropical SST anomalies heavily depends on the sub-tropical and polar front jets [40]. As for the long-term relationship, the SON ASL is positively correlated with the Niño 3.4 index with values of 0.35 in the observations (over 43 years) and 0.47 in CESM2-WACCM (over 165 years) (figure 6(e)), which further validates the teleconnections between La Niña and a deepened ASL.

Zoom In Zoom Out Reset image size

Another factor contributing to the strong ASL is the stratospheric polar vortex, which is strongly coupled with the ASL in austral spring [33, 41]. The year 2021 has a very strong and stable polar vortex (figure 6(c)), which is also reflected in the very low Antarctic ozone in austral spring [42]. The generally weakening of the polar night jet allows for the stratosphere-troposphere coupling through planetary waves [33]. The deepened ASL weakened the planetary wave-1 forcing leads to a stronger polar vortex and later polar vortex breakup [43]. Through stratosphere-troposphere coupling, this is associated with the stronger westerly wind anomalies in the lower troposphere, contributing to the stronger ASL through its interaction with the topography [44]. In contrast, the year 2002 shows the only major sudden stratospheric warming in the SH, relating to a very weak stratospheric polar vortex (figure 6(d)). Their long-term relationship is shown in figure 6, with the correlation between the SON ASL and PCH50 up to 0.74 in observations and 0.67 in CESM2-WACCM (165 years). The correlations are not sensitive to the definition of polar vortex strength (not shown). Based on the above evidence, it is proposed that La Niña conditions in 2021, along with the strong stratospheric polar vortex favoured the occurrence of the very strong ASL.