Abstract. There is an observed relationship linking Arctic sea ice conditions in autumn to mid-latitude weather the following winter. Of interest in this study is a hypothesized stratospheric pathway whereby reduced sea ice in the Barents and Kara seas enhances upward wave activity and wave-breaking in the stratosphere, leading to a weakening of the polar vortex and a transition of the North Atlantic Oscillation (NAO) to its negative phase. The Causal Effect Networks (CEN) framework is used to explore the stratospheric pathway between late autumn Barents–Kara sea ice and the February NAO, focusing on its seasonal evolution, timescale dependence, and robustness. Results indicate that the pathway is statistically detectable and has been relatively active over the 39-year observational period used here, explaining approximately 26 % of the interannual variability in the February NAO. However, a bootstrap-based resampling test reveals that the pathway is highly intermittent: the full stratospheric pathway appears in only 16 % of the sample populations derived from observations, with individual causal linkages ranging from 46 % to 84 % in occurrence rates. The pathway's intermittency is consistent with the weak signal-to-noise ratio of the atmospheric response to Arctic sea ice variability in modelling experiments and suggests that Arctic–mid-latitude teleconnections might be favoured in certain background states. On shorter timescales, the CEN detects two-way interactions between Barents–Kara sea ice and the mid-latitude circulation that indicate a role for synoptic variability associated with blocking over the Urals region and moist air intrusions from the Euro-Atlantic sector. This synoptic variability has the potential to interfere with the stratospheric pathway, thereby contributing to its intermittency. This study helps quantify the robustness of causal linkages within the stratospheric pathway, and provides insight into which linkages are most subject to sampling issues within the relatively short observational record. Overall, the results should help guide the analysis and design of ensemble modelling experiments required to improve physical understanding of Arctic–mid-latitude teleconnections.

Arctic sea ice extent in autumn is significantly correlated with the winter North Atlantic Oscillation (NAO) in the satellite era. However, questions about the robustness and reproducibility of the relationship persist. Here, we show that climate models are able to simulate periods of strong ice-NAO correlation, albeit rarely. Furthermore, we show that the winter circulation signals during these periods are consistent with observations and not driven by sea ice. We do so by interrogating a multimodel ensemble for the specific time scale of interest, thereby illuminating the dynamics that produce large spread in the ice-NAO relationship. Our results support the importance of internal variability over sea ice but go further in showing that the mechanism behind strong ice-NAO correlations, when they occur, is similar in longer observational records and models. Rather than sea ice, circulation anomalies over the Urals emerge as a decisive precursor to the winter NAO signal.

Abstract. It is now well established that the Arctic is warming at a faster rate than the global average. This warming, which has been accompanied by a dramatic decline in sea ice, has been linked to cooling over the Eurasian subcontinent over recent decades, most dramatically during the period 1998–2012. This is a counter-intuitive impact under global warming given that land regions should warm more than ocean (and the global average). Some studies have proposed a causal teleconnection from Arctic sea-ice retreat to Eurasian wintertime cooling; other studies argue that Eurasian cooling is mainly driven by internal variability. Overall, there is an impression of strong disagreement between those holding the “ice-driven” versus “internal variability” viewpoints. Here, we offer an alternative framing showing that the sea ice and internal variability views can be compatible. Key to this is viewing Eurasian cooling through the lens of dynamics (linked primarily to internal variability with some potential contribution from sea ice; cools Eurasia) and thermodynamics (linked to sea-ice retreat; warms Eurasia). This approach, combined with recognition that there is uncertainty in the hypothesized mechanisms themselves, allows both viewpoints (and others) to co-exist and contribute to our understanding of Eurasian cooling. A simple autoregressive model shows that Eurasian cooling of this magnitude is consistent with internal variability, with some periods exhibiting stronger cooling than others, either by chance or by forced changes. Rather than posit a “yes-or-no” causal relationship between sea ice and Eurasian cooling, a more constructive way forward is to consider whether the cooling trend was more likely given the observed sea-ice loss, as well as other sources of low-frequency variability. Taken in this way both sea ice and internal variability are factors that affect the likelihood of strong regional cooling in the presence of ongoing global warming.