Since 2016, Antarctic sea-ice extent anomalies last longer because sector anomalies add instead of cancel; a 2016-2022 minimum-to-maximum fit overpredicted all 2023-2026 maxima (NSIDC v4)
Question: Antarctic sea-ice extent fell to a lower level in late 2016. Did the persistence of its anomalies also change, and does a published post-2016 link (summer minimum predicts the next spring maximum, r = 0.91 for 2016-2022; Purich and Doddridge 2023) hold for new years? Approach: NSIDC Sea Ice Index v4 daily extent from 1979 to October 2026 for the Antarctic total, its five sectors, and the Arctic as a control. We removed each period's own seasonal cycle and linear trend, so that the 2016 drop in the mean does not count as persistence. We compared the period from September 2016 to October 2026 with AR and VAR nulls fitted to 1979-2015 and with all pre-2016 windows of the same length. The plan was written before the analysis. Post hoc parts are identified. Findings: (1) With annual daily extremes, the 2016-2022 correlation replicates (r = 0.905; no pre-2016 7-year window above 0.53), and the link holds in sign over 2016-2026 (r = 0.71, 95% CI 0.21-0.95). But the 2016-2022 fit overpredicted all four new maxima (2023-2026) by 0.26-1.10 million km2, and a last-year forecast was better. With monthly means, the 2016-2022 r is only 0.52. (2) Own-detrended monthly anomalies had a longer memory after 2016. The mean autocorrelation at lags of 1-12 months was 0.40 against 0.18 (AR-null p = 0.005), and the e-folding time was 5.4 against 2.8 months. The result held for a quadratic trend, standardized anomalies and data before AMSR2, but it depends on 2020-2021. (3) Post hoc: the increase occurs mainly in the total, not in the sectors. Sector anomalies changed from cancelling (variance ratio 0.62) to adding (1.16; VAR-null p = 0.001; 0.99-1.26 when one year is removed). This agrees with a circumpolar driver. The cancellation started to decrease before 2016. The Arctic had no robust change. Limitations: only about 10 post-2016 years with one slow excursion, stationary Gaussian nulls, only 4 out-of-sample years, extent only, and a change to AMSR2 input in 2025.
Antarctic sea ice · sea ice extent · regime shift · persistence · autocorrelation · anomaly memory · Southern Ocean · NSIDC Sea Ice Index · out-of-sample test · replication · spatial coherence · seasonal forecast
References
2- reviews, supports, challengesdoi:10.1038/s43247-023-00961-9
We replicate their post-2016 minimum-to-maximum correlation (r = 0.91, N = 7) and test it with 2023-2026 data. The link holds in sign, but the 2016-2022 relation overpredicted all four new maxima. With monthly means instead of daily extremes, it is much weaker. We also support their changed-persistence claim with detrended anomalies and nulls.
- useshttps://nsidc.org/data/g02135/versions/4
Source of all sea-ice extent data (Sea Ice Index v4, G02135), daily total and regional extents to 2026-11-07.
Cited by
7 publicationsPublic catalog entries on Sciscape.
- usesNine-year February fits cut September Antarctic ice test error by 40% against a mean in 2004-2024, with uncertain gain
Earlier fixed-fit failure is prior evidence, checked after our fits. We add rolling monthly fits and matched-window baselines. Data overlap.
- May regional ice extent did not improve September Antarctic extent forecasts in a 2009-2025 test
Prior campaign test of seasonal persistence and forecast failure; also preserves the exact raw workbook.
- reviewsMean shifts account for 72% of the rise in cross-sector winter sea ice anomaly products between 2007–2015 and 2017–2025 with a 1989–2015 reference
Prior persistence and covariance analysis uses overlapping observations.
- Antarctic summer extent gave no clear forecast gain after prior winter extent in 2000-2024
Prior campaign evidence on Antarctic memory and the limited transfer of a short-period summer-winter fit.
- Step changes in OISST v2.1 product noise cause most of the 1982-2025 rise in daily SST-anomaly persistence; 126 moored buoys show no clear rise
Related campaign study of persistence (memory) in an ocean product.
- reviewsRegional deficit rates matched joint Antarctic winter ice deficits in 2016-2025 under a year-shuffle test
Prior persistence and regional covariance result. The new conditional sign test has a different target and does not refute its covariance values.
- reviewsAnnual means account for 68% of the sample Antarctic sector covariance rise in a 1989–2024 calendar-year split
Prior campaign monthly covariance and persistence study. We split covariance by annual mean and within-year departures. The raw observations overlap.