the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 21 Apr 2026
Brief Communication: The Interdecadal Pacific Oscillation is not a Reliable Indicator of the Timing of an Ice-Free Arctic
Abstract. The Arctic is transitioning towards a seasonally ice-free state over the next few decades, with internal variability determining precisely when. The phase of the Interdecadal Pacific Oscillation (IPO) has been proposed to reduce internal variability uncertainty in ice-free predictions. Using a selection of 8 large ensembles from the Coupled Model Intercomparison Project - Phase 5 (CMIP5) and CMIP6, starting from current mean sea ice extent, we find no agreement that the IPO affects ice-free timing. However, a majority of large ensembles indicate the IPO can affect the rate and spatial distribution of Arctic sea ice decline over the next two decades.
- Preprint
(31399 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 11 Jun 2026)
- RC1: 'Comment on egusphere-2026-2082', Anonymous Referee #1, 05 May 2026 reply
-
RC2: 'Comment on egusphere-2026-2082', Anonymous Referee #2, 05 Jun 2026
reply
This Brief Communication presents a relevant contribution to the study of Arctic sea-ice climate variability. In particular, the authors investigate whether the relationship between the current phase of the Interdecadal Pacific Oscillation (IPO) and the timing of an ice-free Arctic Ocean, previously identified in CESM1 projections by Screen and Deser (2019), is robust across other models. They extend the analysis to multiple large ensembles from CMIP5 and CMIP6 and find no consistent agreement across models that the initial IPO phase affects the timing of an ice-free Arctic Ocean. Nevertheless, in the majority of large ensembles, they do find evidence that the initial IPO phase significantly affects the sea-ice extent trajectory, although at different lead times across models.
The manuscript is well written and presents an interesting analysis that contributes to a better understanding of Arctic sea-ice variability. However, a few points should be clarified before publication.
- Lines 54–56: Screen and Deser (2019) use an 11-year running mean for the IPO. Please clarify whether the same approach is used in the present study.
- Pooling experiment methodology: Please clarify the methodology used for the pooled experiments. In Figure 2, all trajectories appear to be initialized at 4.7 million km². Does the pooling procedure replace each ensemble member with an average over shifted versions of itself, or are the shifted trajectories treated as separate samples?
- Figure 2: In panels A, there are small grey boxes near the left-hand side of the plots. Please indicate in the caption what these represent, or remove them if they are not needed. Please also clarify what the hatched regions indicate in panels B–E.
- Section 3.2 methodology: Please clarify how the analysis in Section 3.2 is performed. Are the same trajectories as in Section 3.1 used? If so, at a given lead time before the ensemble-mean ice-free year, do different members start from different initial sea-ice extents?
- Line 115: Please clarify whether this sentence refers to the scatterplots in Figure 3 panels I–P.
Citation: https://doi.org/10.5194/egusphere-2026-2082-RC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 210 | 77 | 24 | 311 | 19 | 16 |
- HTML: 210
- PDF: 77
- XML: 24
- Total: 311
- BibTeX: 19
- EndNote: 16
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
This short paper is an important contribution. It takes as a starting point the work of Screen and Deser, which suggested with a single model that the IPO could be a useful constraint on the timing of an ice-free Arctic. As with all such potential constraints, it is valuable to test their robustness across multiple models. This submission tests the constraint across 8 models and finds that it is not robust. I support publication after revision and/or suitable responses to the following points:
Line 60: I don’t understand the rationale for the ensemble pooling. The text states “to incorporate multiple initial conditions”, however, as far as I understand it, these are not independent runs but multiple samples from the same ensemble member. Unlike a distinct ensemble member, which is free to evolve differently to all other members, these pooled samples are not independent and are simply offset by a few years. This process seems to artificially inflate the ensemble size, without providing independent samples, and needs better justification.
Line 76: Does this increase in the percentage of LEs showing significant divergence come from the alternative IPO definition and/or the pooling of initial states? If the latter, I’m concerned that significance is inflated artificially and has limited physical meaning. Does the significance testing assume each sample is independent? They aren’t (see above). Besides, regardless of the significance or not of the divergence, the more important point seems to be that the divergence is small around the time of first ice-free conditions, which is independent of the pooling (as the samples still have the same timing of ice-free conditions). Basically, I’m wondering if the pooling is necessary, even if it could be justified.
Section 3.2: I think this section could be clearer. There are lots of kind-of “what if” statements and the key result – which I think is it that there is no consistent lagged relationship between IPO and ice-free conditions across models (in fact, in most models there is no relationship at all) – gets a bit lost. I think this point could be made more succinctly and clearly, alongside some simplification of Figure 3 (see comment below).
Figure 3: It’s unclear what value panels I-P add, and they are not directly referred to in the text. If I understand correctly, the correlation shown in panels I-P can already been inferred from the blue lines in panels A-H. Panels A-H include a second metric for ice-free, which is not used (or introduced) before. Is this necessary? What is the added value of the red line? In such a short note, I’m in favour of parsimony: keeping things as simple as possible.
Conclusions: The lack of robustness of the constraint across models is an important result, but there is scope for deeper consideration of the causes of this lack of robustness, its implications, and potential ways forward. The IPO constraint can only ever reduce the model uncertainty due to internal variability. Its utility is dependent on the timeframe for an ice-free Arctic, which also depends on the simulated forced response. In this regard, it might not be appropriate to consider all models as equal. There is no effort to critique if any of models have (un)realistic forced responses, for example, by looking at their historical trends or sea-ice sensitivity (sea-ice loss per degree of global warming). At the very least, this requires some thought and discussion. A more sophisticated approach might be to first constrain the model spread in the forced response, before considering if internal variability can additionally constrain the spread around this forced response.
Other minor points:
Typo line 18: “showed the IPO has been shown”
Figure 1: might be helpful to list the number of members on the titles
Line 134: “at today’s lead times” confusing and probably not needed. The sentence makes (more) sense without this bit.