Nonstationary Relationship Between Autumn Arctic Sea Ice and the Winter North Atlantic Oscillation

The North Atlantic Oscillation (NAO) has a dominating influence on wintertime weather in the North Atlantic region, and therefore, it is of great interest to predict the NAO several months ahead. While state‐of‐the‐art dynamical forecast models appear to be increasingly skillful in predicting the NAO, statistical methods with comparable or higher predictive skill are still often used. An inherent problem with statistical methods is that any empirical relationship between predictors and the NAO may be valid for some periods but subject to change over time. Here we use a set of new centennial reanalyses and large‐ensemble simulations with multiple climate models to discover clear evidence of nonstationarity in the lagged correlation between autumn Barents‐Kara sea ice and the winter NAO. This nonstationarity leads us to question the causality and/or robustness of the ice‐NAO link. We caution against indiscriminately using Barents‐Kara sea ice to predict the NAO. Plain Language Summary European winter weather is heavily influenced by the North Atlantic Oscillation (NAO). The so‐called positive NAO brings mild and wet conditions to northern Europe in winter, and the negative NAO tends to be cold and dry. Scientists attempt to forecast the NAO in advance by one of two ways: using complex weather forecast models or using relatively simple statistical equations. Although statistical methods can outperform more complicated forecast models, they assume that predictor relationships do not change over time. This assumption is not always valid. In this study we examined the relationship over time between autumn sea ice in the Barents‐Kara Seas and the winter NAO. In recent decades, a strong relationship has been observed whereby especially reduced autumn sea ice often precedes negative NAO in the following winter. When we looked further back in time, however, we found that the ice‐NAO relationship has been highly changeable and sometimes, the complete opposite to that seen recently. An analysis of hundreds of simulations from multiple climate models confirms that the ice‐NAO relationship varies a lot, just due to natural climate variability. Our results suggest it is unwise to make predictions of the winter NAO based on autumn sea ice. Key Points The observed correlation between autumn Barents‐Kara sea ice and the winter NAO has not been stationary over time An ice‐NAO correlation as high as in recent decades is rare since 1865 but within the possible range due to int