Antarctic extreme seasons under 20th and 21st century climate change

Projected 21st-century changes in climate over Antarctica and the Southern Ocean are expected to have major impacts on physical and biological systems, with global implications. While time-mean Antarctic climate responses to greenhouse gases and stratospheric ozone changes have been extensively studied, variability and extreme weather and climate events—especially extreme seasons—are of greater importance for impacts because seasonal persistence can produce accumulated effects (e.g., ice shelf surface melt leading to break-up, and impacts on penguin breeding). Key atmospheric variables include near-surface temperature, precipitation, and near-surface wind. A knowledge gap exists on how extreme seasons may change over Antarctica and the Southern Ocean under future forcing scenarios; Antarctica was not included in the IPCC AR6 WG1 Chapter 11 on extremes, partly because rigorous assessments require large ensembles that have only recently become widely available. Changes in seasonal extremes need not follow mean-state changes; for example, sea ice strongly influences year-to-year surface air temperature variability and seasonal extremes by limiting very low temperatures when ice-free, and reduced meridional temperature gradients can decrease synoptic variability. The Southern Hemisphere mid-latitude westerly jet shifts poleward and strengthens under greenhouse gas increases and ozone depletion (especially in summer), affecting variability and precipitation as storm tracks move with the jet. Extreme seasonal winds are closely associated with seasonal mean wind, implying potential changes with jet shifts. This study uses CMIP6 large ensembles to provide a broad overview of extreme seasons for near-surface temperature, precipitation, and near-surface westerly winds over Antarctica and the Southern Ocean. Extremes are defined relative to an evolving background climate (not fixed thresholds). The key research questions are: • How will extreme warm and cold seasons evolve into the future, especially in regions of retreating sea ice? • How will the projected poleward shift and strengthening of circumpolar tropospheric westerlies influence changes in seasonal extremes?

The paper situates its work within established findings: (1) Sea ice exerts a strong control on interannual variability and extremes of surface air temperature by limiting the occurrence of very low temperatures; observed warming and loss of cold extremes in the Antarctic Peninsula have been linked to increased ice-free ocean conditions. (2) Polar amplification reduces meridional temperature gradients, contributing to reduced synoptic temperature variability. (3) The Southern Hemisphere mid-latitude westerly jet has shifted poleward and strengthened under greenhouse gas forcing and ozone depletion, with associated storm track changes and high-latitude precipitation increases; seasonal mean winds correlate with seasonal extremes. (4) Increases in extreme precipitation are expected with warming due to higher atmospheric moisture capacity, modulated by dynamical changes (e.g., storm track shifts) and regional thermodynamic factors (e.g., sea ice retreat). Prior CMIP studies identified model sensitivity to sea ice biases and jet representation, affecting projections of Antarctic temperature, precipitation, and winds. The study leverages these insights to interpret differences between changes in background climate and changes in seasonal extremes across variables and seasons.

Data: The study uses CMIP6 large ensemble (LE) simulations from five models with many realisations: ACCESS-ESM1-5 (40), CanESM5 (50), MIROC6 (50), MPI-ESM1-2-LR (30), and UKESM1-0-LL (14). Historical simulations are used up to 2014 and SSP3-7.0 from 2015–2100. Variables analyzed are monthly means aggregated to seasonal means for austral summer (DJF) and winter (JJA): 2 m surface air temperature (tas, TAS), 10 m zonal wind (uas, UAS), and total surface precipitation rate (pr, PR). Sea ice concentration (siconc) is used to derive a sea ice equivalent latitude (SIE-equivalent) diagnostic. ERA5 reanalysis provides observed-evolution comparisons for the same variables over 1979–2023, spatially averaged over defined regions for compatibility with the coarser model grids. Sea ice equivalent latitude: For each model, the equivalent latitude of the sea ice edge is computed from monthly SIE (first realisation only) and the area south of 60°S (ocean area from areacello). The equivalent latitude conceptualizes the zonal-mean latitude to which sea ice would extend if zonally symmetric. Climatologies are computed over relevant 31-year periods; an ice-free Antarctic corresponds to an equivalent latitude of ~71°S. A multi-model mean equivalent latitude is used for zonal-mean figures. Regridding and regional means: For multi-model spatial maps, data from each model are interpolated using linear regridding to a common grid (the CanESM5 grid, the coarsest among the models). Regional averages are defined for the northern Antarctic Peninsula (62.5°S–67.5°S, 70°W–57.5°W) and a Southern Ocean sector (62.5°S–67.5°S, 175°E–175°W) to compare CMIP6 outputs with ERA5. Definition of background climate and extremes: For each model m, grid point x, season s, and realisation r, the seasonal time series X_t^(x,m,r) (1850–2100; historical joined to SSP) is decomposed as X_t^(x,m,r) = X̄_t^(x,m) + ε_t^(s,x,m), where X̄_t^(x,m) is the background climate and ε are residuals. The background climate is computed as the mean over all realisations at time t (realisation mean) followed by Lanczos low-pass filtering with a 31-year moving window and a 20-year cutoff period to remove remaining high-frequency variability in models with smaller ensemble sizes. Tests with 10- and 5-year cutoffs show similar results. Extreme thresholds are defined relative to the evolving background climate using residuals ε from all realisations within a 31-year moving window centered on year t (t−15 to t+15). The 10th percentile (p10) and 90th percentile (p90) of ε are computed for each model, location, and season; the extreme seasonal values are then X̂_p10(t) = X̄_t + p10(ε) and X̂_p90(t) = X̄_t + p90(ε). For example, with CanESM5’s 50 members, each year’s percentile estimates use 1550 residuals (31 years × 50 realisations). Multi-model means of background climate, p10, and p90 are computed as unweighted averages across models. Diagnostics and periods: Spatial maps and zonal means compare 1960 (pre-ozone-hole baseline) and late-century (2080) states, and differences (2080–1960). Time series at example regions illustrate evolutions and compare with ERA5. Figures assess temperature (TAS), near-surface westerly winds (UAS), and precipitation (PR) extremes and their range (p90−p10), along with SIE-equivalent latitude. The last plotted model year is 2085 because of the 31-year smoothing window.

- Extremes vs mean-state changes: Externally forced changes in extreme seasons generally do not mirror changes in the background climate, and the nature of differences varies by variable (TAS, UAS, PR) and season (DJF, JJA). - Temperature (TAS): In winter (JJA), regions of sea ice retreat show larger warming in cold extremes (p10) than in warm extremes (p90) and background means, extending into adjacent coastal margins (notably the northern Antarctic Peninsula). Zonal mean differences between p90 and p10 changes peak over the sea-ice zone, reaching about 1.74 °C at 63°S. In the northern Antarctic Peninsula, 1960–2080 winter warming is 5.6 °C for TAS_p10 versus 4.4 °C for TAS_p90. In summer (DJF), changes in extremes closely follow background warming, with smaller internal variability (narrower p90–p10), so projected changes can shift conditions comparably (or more) beyond the historical range despite smaller absolute warming. - Near-surface westerly wind (UAS): Summer background changes show strengthening and poleward shifts of the circumpolar westerlies and weakening coastal easterlies, with pronounced increases during the late-20th-century ozone depletion period (notably the 1980s). Extremes broadly follow background changes in summer, but the peak in UAS variability (p90−p10) near ~60°S shifts poleward with the jet, causing a reduction in variability between ~55–60°S up to about 1.2 m s−1. At a midlatitude Southern Ocean site (175°–185°), p10 increases by ~2.0 m s−1 (1960–2080), larger than background (~1.44 m s−1) and p90 (~0.75 m s−1). In winter, zonal-mean jet structure is diffuse, and changes in extremes largely mirror background changes with little impact on seasonal variability. ERA5 time series show similar variability ranges but are a single realisation, complicating trend comparisons. - Precipitation (PR): Background precipitation increases broadly over the Southern Ocean and regionally over coastal Antarctica, especially the western Antarctic Peninsula. High extremes (p90) generally increase more than low extremes (p10) at most locations. Zonal-mean PR changes resemble those of UAS (single summer maximum near ~60°S; broader winter increases), but the distinct summer narrowing in UAS variability at 55–60°S is not clearly mirrored in PR extremes. Enhanced increases in high PR extremes occur along coastal Antarctica and the Peninsula (~60–70°S) in both seasons. ERA5 shows seasonal differences with summer increases during the ozone hole formation period and large interannual variability. - Mechanisms: Winter TAS extreme differences are linked to sea ice retreat reducing the frequency of very cold seasons; summer UAS/PR changes relate to jet/storm track shifts and moisture availability. Model examples underscore sea-ice roles: UKESM1-0-LL (with substantial sea ice retreat) shows pronounced winter p90−p10 differences near the sea-ice edge; MIROC6 (little sea ice) shows negligible differences. - Robustness and model spread: The qualitative patterns are robust across five LE models. Differences arise where models have biases (e.g., MIROC6, MPI-ESM1-2-LR with negative sea-ice biases and equatorward winter jet biases), or differing ozone chemistry representation (UKESM1-0-LL with interactive ozone exhibits weaker summer westerly strengthening during ozone recovery).

Findings demonstrate that Antarctic seasonal extremes respond differently from the mean climate, with mechanisms varying by variable and season. For winter TAS, sea ice retreat curtails the occurrence of extremely cold seasonal means, leading to stronger warming of cold extremes than warm extremes and background states, including over adjacent coastal regions. In summer, TAS extremes closely track background warming because newly exposed ocean is not substantially warmer than the atmosphere and summer sea-ice extent is smaller, limiting thermodynamic amplification. For UAS, extremes and background changes are tightly coupled to the jet and storm tracks: the poleward shift moves the zone of maximum seasonal wind variability, reducing variability at 55–60°S in summer. Winter jets are more diffuse and zonally asymmetric, leading to smaller changes in variability and closer mirroring of background changes. For PR, high extremes increase slightly more than low extremes and background, consistent with thermodynamic moisture increases, but spatial patterns and magnitude are modulated by dynamical factors (jet/storm track shifts) and regional thermodynamics (sea ice retreat and open-ocean expansion). Model sensitivity analyses show that these conclusions are qualitatively robust, though models with sea-ice and jet biases exhibit weaker high-latitude PR extreme increases, and interactive ozone chemistry affects summer jet responses during ozone recovery. The implications include earlier emergence of impactful changes in summer TAS relative to historical variability, with potential consequences for ice shelf stability and ecosystems, and a projected reduction in summer wind variability on the jet’s poleward flank that may influence Southern Ocean carbon sink variability. Further work is needed to disentangle contributions of sea ice versus SST changes to PR extremes, to link sub-seasonal events to seasonal extremes, and to address dynamical questions such as seasonal contrasts in variability on the jet’s poleward side (e.g., roles of cyclonic wave breaking and eddy feedbacks).

This study delivers the first multi-variate assessment of Antarctic seasonal extremes across TAS, UAS, and PR using CMIP6 large ensembles, defining extremes relative to an evolving background climate. Key contributions are: (1) extremes generally diverge from mean-state changes, with variable- and season-dependent behavior; (2) winter TAS exhibits larger warming in cold extremes in regions of sea-ice retreat, reducing the frequency of extreme cold seasons; (3) summer UAS extremes reflect jet shifts, with a poleward migration of maximum variability and reduced variability at 55–60°S; (4) PR high extremes increase slightly more than background and low extremes, influenced by jet structure and possibly sea-ice retreat. These conclusions are qualitatively robust across models, though biases in sea ice and jet latitude modulate magnitudes, and interactive ozone chemistry affects summer jet evolution. Impacts-related implications include earlier emergence of significant summer TAS changes relative to historical variability and potential effects of reduced summer wind variability on the Southern Ocean carbon sink. Future research should target sub-seasonal extremes and their contribution to seasonal anomalies, moisture-source attribution for PR extremes (e.g., via sensitivity studies or moisture tracing), improved representation of jet dynamics (including wave breaking and eddy feedbacks), tropical teleconnections, and downscaling to resolve local processes (e.g., foehn winds) that influence Antarctic extremes.

- Model biases: Some ensembles (e.g., MIROC6, MPI-ESM1-2-LR) exhibit negative sea-ice biases and equatorward winter jet biases, reducing capacity for high-latitude open-ocean expansion and altering storm track interactions with Antarctica, which can weaken projected increases in high PR extremes. - Ozone representation: Only UKESM1-0-LL includes interactive ozone chemistry, leading to different summer jet responses during ozone recovery relative to fixed-ozone models; this introduces model spread in UAS trends during late 20th to 21st century. - Resolution: Large-ensemble configurations are necessarily low resolution, limiting the representation of local processes (e.g., foehn winds) and detailed storm dynamics; downscaling is needed for impacts-relevant detail. - Attribution complexity: Disentangling thermodynamic (moisture availability, sea-ice retreat) from dynamical (jet/ storm track) contributions to PR extremes is challenging without targeted sensitivity experiments or moisture-tracing diagnostics. - Observational comparison: ERA5 provides only a single realisation, complicating direct comparison of ensemble mean trends; large internal variability particularly affects UAS and PR. - Methodological constraints: The analysis uses regridding to a coarse common grid and multi-model unweighted means, which may mask model-specific spatial features; the extreme definitions rely on 31-year windows and ensemble residuals, which may smooth shorter-term variability and depend on ensemble size.