Who causes whom? A spatially distributed causal analysis of the relationship between Arctic sea ice and teleconnection indices

The relationship between Arctic sea ice and tropical climate variability is a crucial aspect of global climate dynamics. While numerous studies have explored potential links between sea ice concentration (SIC) or sea ice thickness (SIT) and teleconnection indices such as AMO, AO, NAO, ENSO, and PDO, these investigations often faced challenges in fully capturing the complexity of these interactions. For instance, most analyses relied on linear, non-causal methods such as trend matching (although the underlying processes are likely highly nonlinear), or focused on single indices (thus potentially missing more complex interactions when more than one index is considered at once), or analyzed the relationship in aggregate over the entire Arctic region, rather than considering subtle regional differences. Additionally, these teleconnections were often assessed in only one “direction” (e.g., how much ENSO influences SIC), but there is evidence to suggest that there may be two-way interactions at play.

In this study, we address these challenges by proposing a bi-directional, causal, and spatially distributed approach to analyze the relationships between SIC/SIT and eight teleconnection indices. Using transfer entropy (TE), a non-parametric measure of information flow, we quantify the influence of these indices on SIC/SIT and vice versa across multiple lead times. This approach lets us understand how these causal relationships vary at different lead times and over different Arctic regions, to verify whether the various teleconnection indices provide information that is complementary or redundant, and to detect preferential directions in the causal relationship between indices and ice (thus answering the question “who influences whom?”). For instance, our results indicate that the North Atlantic Oscillation is influenced by the Arctic ice more than it itself affects the ice, whereas the relationship is inverted for the Atlantic Multidecadal Oscillation.

Although we focus our analysis on understanding the spatial and temporal variability of Arctic-teleconnection interactions, the proposed framework is highly flexible and can be adapted to consider other indices and lead times, and entirely different domains altogether.