Early prediction of extreme stratospheric polar vortex states based on causal precursors

Variability in the stratospheric polar vortex (SPV) can influence the tropospheric circulation and thereby winter weather. Early predictions of extreme SPV states are thus important to improve forecasts of winter weather including cold spells. However, dynamical models are usually restricted in lead time because they poorly capture low-frequency processes. Empirical models often suffer from overfitting problems as the relevant physical processes and time lags are often not well understood. Here we introduce a novel empirical prediction method by uniting a response-guided community detection scheme with a causal discovery algorithm. This way, we objectively identify causal precursors of the SPV at subseasonal lead times and find them to be in good agreement with known physical drivers. A linear regression prediction model based on the causal precursors can explain most SPV variability (r² = 0.58), and our scheme correctly predicts 58% (46%) of extremely weak SPV states for lead times of 1–15 (16–30) days with false-alarm rates of only approximately 5%. Our method can be applied to any variable relevant for (sub)seasonal weather forecasts and could thus help improving long-lead predictions.