Sub-seasonal Arctic-midlatitude linkages as causal networks

<p>Arctic-midlatitude linkages are state-dependent and multi-facetted. Their faithful quantification on sub-seasonal time scales therefore requires methods that account for differences in the background state and reject spurious links arising from common drivers and strong autocorrelation. Here, we use causal network discovery methods that have these properties to establish causal linkages of sub-seasonal time series of selected atmospheric circulation and surface condition indices in the Arctic and Northern mid-latitudes. In order to obtain state-dependent and uncertainty-quantified linkages, we use a large set of sub-seasonal reforecasts from the ECMWF model cycle 47R1 with 11-member ensembles initialised on each day of the winters 1999/2000 to 2018/2019. Pooling forecasts started within a 10-day window allows enough samples to robustly estimate a causal-effect network valid for the window. State-dependence and uncertainty quantification of Arctic-midlatitude linkages is then possible by analysing he seasonal and year-to-year changes of the networks. As an application of the general methods, we discuss the causal links diagnosed for Northern Europe surface air temperature in winter. We chose this index because of its societal relevance and previously reported strong links between extreme cold winters in this region and Arctic sea ice anomalies. Somewhat surprisingly, on sub-seasonal time scales we do not detect significant links between Northern Europe surface air temperatures and Arctic sea ice indices, nor with the polar vortex strength. However, the North Atlantic Oscillation index and geopotential height at 500 hPa over the polar cap do have causal links to Northern Europe surface air temperatures as expected. Link strengths have a seasonal dependence, as well as substantial year-to-year changes. Our results demonstrate the value of large forecast ensembles for quantifying explicitly the state-dependence and uncertainty of Arctic-midlatitude linkages, and might provide guidance for reconciling some previously reported conflicting results.</p>