Record low Antarctic sea ice coverage indicates a new sea ice state

Antarctic sea ice is a critical component of the global climate system. The study addresses why Antarctic sea ice reached unprecedented lows in 2023—part of a cluster of record-low summers since 2016—and whether these extremes represent a shift to a new sea ice state. The authors highlight that sea ice coverage in January–February 2023 set monthly and daily records for low extent, with exceptionally low coverage persisting into autumn and winter. They investigate whether recent extremes are manifestations of interannual variability or indicative of a statistically distinct regime, and assess the potential role of Southern Ocean subsurface warming relative to atmospheric climate modes traditionally linked to Antarctic sea ice variability.

Previous work documented a multidecadal Antarctic sea ice increase peaking around 2014, followed by a sharp decline in 2016. Suggested contributors to earlier high sea ice included strengthened high-latitude winds and Antarctic meltwater inputs, with ocean–sea ice feedbacks potentially sustaining high coverage. Individual low-ice events have been tied to anomalous atmospheric circulation (e.g., zonal wave number three in 2016; Amundsen Sea Low in 2021/22). However, large-scale modes (SAM, ENSO, IPO, AMO) have predominantly regional impacts and their phases do not align with observed recent regime changes. The paper situates findings within hypotheses of a two-timescale Southern Ocean response to strengthened westerlies (initial surface cooling then longer-term subsurface warming) and discusses uncertainties around the role of accelerated Antarctic meltwater, which could freshen and stabilize the near-surface ocean, potentially increasing sea ice, yet current observations show declining sea ice concurrent with subsurface warming.

Data and indices: National Snow and Ice Data Center (NSIDC) Southern Hemisphere sea ice extent (SIE) indices (daily, five-day, monthly; Nov 1979–Jun 2023) were used. Anomalies were computed relative to 1979–2022 climatologies (daily and five-day also against their respective climatologies). Missing December 1987 and January 1988 monthly values were interpolated from adjacent months following Simmonds. To contextualize late-summer extremes, monthly SIE anomalies were also expressed as percentages of the 1979–2022 monthly climatology. Change-point detection: An algorithm (R package changepoint; method PELT; BIC penalty) identified mean shifts in the monthly SIE anomaly time series, yielding two change points (August 2007 and August 2016), partitioning the record into three periods. Robustness was tested with strucchange (BIC-optimal with h=0.15), which identified three change points (Aug 1993, Nov 2007, Aug 2016) for linear model structures; when constrained to two, it also selected Aug 2007 and Aug 2016. Variances and means across periods were compared using F-tests and t-tests (Welch’s for unequal variances between the first and later periods; equal-variance t-test between the latter two), showing statistically distinguishable means and lower variance in the earliest period. Climate mode analysis: Time series for SAM, Southern Oscillation Index, Interdecadal Pacific Oscillation, Indian Ocean Dipole, and Atlantic Multidecadal Oscillation were obtained, smoothed with a centered two-year rolling mean, values near zero masked (±0.5 SD), and sign changes compared against SIE change points. No alignment was found. Ocean temperature: Gridded Argo product (1°×1° monthly; surface to 2000 dbar; Jan 2004–May 2023) provided temperature anomalies (relative to 2004–2022). Circumpolar averages were calculated over 50–65°S (grid-aware via xgcm). Depth–time sections and zonal means were derived; linear trends at each depth were tested (significant at all depths, p<2.5e−6; ~0.09 °C decade−1 near surface; ~0.03 °C decade−1 at 500 m). Limitations include sparser early Argo sampling and limited coverage within the seasonal ice zone (data primarily north of sea ice extent). Sea ice concentration and longitudinal extent: NSIDC CDR v4 monthly SIC (Nov 1978–May 2023; Jan–May 2023 NRT v2) were regridded to 1°×1°. For each longitude, longitudinal SIE was computed by summing area of grid cells with SIC>15%. Anomalies were relative to 1979–2022. Hovmöller and smoothing: Longitudinal SIE and ocean temperature anomalies (0–100 m and 100–200 m, averaged 50–65°S) were smoothed temporally (12-month running mean) and longitudinally (20° running average) to construct Hovmöller diagrams. Composite and significance testing: Spatial composites of SIC and 100–200 m ocean temperature anomalies were compared between periods using Welch’s unequal variance t-tests at each grid point. Bonferroni correction was applied (p≈9×10−6 threshold), acknowledging conservatism due to spatial correlation. Winds and Ekman pumping: ERA5 monthly surface wind stresses (eastward, northward) were used to compute annual means (1979–2022; 2004–2022) and downward Ekman pumping ωe=(∇×τ)/(ρf) with ρ=1025 kg m−3. These contextualize prolonged upwelling conditions. Event metrics: Record daily and monthly SIE minima were quantified against climatologies, including the 19 February 2023 daily minimum and January–June 2023 anomalies.

- Record lows: On 19 Feb 2023, Antarctic SIE reached 1.77 million km², which is 1.02 million km² (36%) below the 1979–2022 average daily minimum; January 2023 monthly mean was 1.74 million km² below average, and February 2023 was 1.12 million km² below average. As percentages, January and February 2023 were −35% and −37% of their climatologies, the largest negative anomalies in the 44-year record. - Persistent deficits: Throughout 2023 to early July, SIE remained at or near record lows for the time of year, with June 2023 showing the largest negative monthly anomaly of the satellite era (−2.33 million km² relative to the 1979–2022 June mean), roughly double the size of the previous record for June. - Regime identification: Change-point analysis of monthly SIE anomalies reveals three statistically distinct periods: Nov 1978–Aug 2007 (lower variability, higher mean), Sep 2007–Aug 2016 (high sea ice state), and Sep 2016–Jun 2023 (low sea ice state). Means are statistically different (t-tests, p<0.01); early-period variance is significantly lower (F-test, p<0.01). - Ocean warming linkage: Circumpolar Southern Ocean subsurface warming (50–65°S) intensified below 100 m from 2015 and at the surface from late 2016. The warming onset precedes the 2016 SIE change point, strongly suggesting subsurface warming contributed to the transition into, and maintenance of, the low ice state. - Spatial coherence: Hovmöller analyses show strong spatial agreement between warm anomalies at 100–200 m (e.g., in 2015 at 340–20°E and 190–250°E) and subsequent negative SIE anomalies in 2016 (10–90°E and 190–290°E), consistent with eastward advection. - Climate modes insufficient: The identified SIE change points do not align with sign changes in SAM, ENSO, IPO, IOD, or AMO. Moreover, the canonical SAM–sea ice relationship has weakened/broken: 2022 and 2023 minima occurred during near-normal surface temperatures and positive SAM, while the subsurface ocean was anomalously warm. - Changed persistence characteristics: Pre-2016, spring maximum SIE correlated with the following summer minimum (1979/80–2015/16 r=0.33, p<0.05; 2007/08–2015/16 r=0.62, p<0.10). Post-2016, this correlation is not significant (2016/17–2021/23 r=0.27, p≈0.5). Conversely, since 2016, summer minimum SIE strongly correlates with the following spring maximum (r=0.91, p<0.01), indicating altered seasonal memory consistent with a stronger role of subsurface ocean conditions. - Ocean trend significance: Linear trends in Argo-era ocean temperature anomalies are significant at all depths (p<2.5×10−6), with magnitudes ~0.09 °C decade−1 near surface and ~0.03 °C decade−1 at 500 m. - Implications: Extremely low sea ice affects ice-shelf basal melt and stability, coastal exposure, dense shelf water production, bottom water formation, deep-ocean ventilation, and penguin colonies; findings underscore urgent need to reduce greenhouse gas emissions.

The analyses indicate Antarctic sea ice has entered a statistically distinct low-extent state since late 2016. The onset of circumpolar subsurface Southern Ocean warming preceding the 2016 transition, and its persistence, provides a mechanistic basis for sustained low ice conditions that is not captured by atmospheric climate modes alone. The breakdown of the conventional SAM–sea ice linkage and the emergence of new persistence characteristics (strong summer-to-following-spring correlation) further support a change in regime where subsurface ocean conditions play a central role in seasonal sea ice evolution. While internal variability and episodic atmospheric anomalies contributed to individual events (e.g., spring 2016), a circumpolar, sustained reduction is difficult to reconcile with regional atmospheric drivers alone. The findings are consistent with a hypothesized two-timescale response of the Southern Ocean to prolonged westerly wind changes (initial surface cooling followed by longer-term subsurface warming), though the exact mechanisms and timing remain debated across models. The role of Antarctic meltwater is uncertain; despite accelerated melt—which could enhance stratification and surface cooling—observations show low sea ice concurrent with subsurface warming. Continued low extents in coming years would strengthen the case for a persistent regime shift. The broader climate implications are substantial, motivating enhanced under-ice observations and targeted process studies to clarify ocean–sea ice feedbacks and improve predictability.

The study identifies a new low-extent Antarctic sea ice state beginning in 2016, characterized by record and persistent deficits culminating in unprecedented anomalies in early 2023. Change-point analysis confirms a distinct regime, and multiple lines of evidence link this shift to intensified subsurface Southern Ocean warming that began prior to the transition. The seasonal persistence of sea ice has changed, with post-2016 conditions showing strong linkage from summer minima to the following spring maxima, suggesting altered ocean–ice feedbacks. While atmospheric variability influences individual events, it does not explain the circumpolar, sustained decline. These results imply that anthropogenic warming of the Southern Ocean is now exerting a dominant influence on Antarctic sea ice, aligning with long-standing model projections of eventual decline. Future research should prioritize expanded under-ice ocean observations, dedicated process studies, and coupled modeling to unravel mechanisms governing the new persistence patterns and to quantify the roles of winds, upwelling, and meltwater. The far-reaching impacts on ice shelves, ocean ventilation, ecosystems, and human activity underscore the urgency of mitigating greenhouse gas emissions.

Key limitations include the relatively short satellite sea ice record (44 years) and the small sample size of the current regime (N=7 years), which constrain statistical power for persistence changes. Under-ice ocean observations remain sparse; the Argo dataset has limited coverage within the seasonal sea ice zone and lower early-year sampling density, potentially biasing early anomalies toward climatology. Spatial significance testing employs a highly conservative Bonferroni correction that may under-identify significant areas due to spatial correlation. Attribution remains uncertain regarding the exact mechanisms (e.g., two-timescale wind response, meltwater impacts, internal variability), and this study does not conduct process-resolving modeling to separate drivers. Some key climate indices may not fully capture regional dynamics, and observed relationships could evolve with additional years of data.