AI reconstruction of European temperature and precipitation anomalies from Euro-Atlantic weather regimes

Euro-Atlantic weather regimes (WRs) provide a description of quasi-stationary large-scale circulation patterns that strongly modulate European weather variability and extremes. Yet, most existing work focuses on the correlation and impacts of the WR on European weather, while the estimation of ground-level meteorological variables, such as temperature and precipitation, from Euro-Atlantic WR remains largely unexplored.

This contribution presents an AI-based framework that maps Euro-Atlantic WR indices to monthly European 2-m temperature and precipitation anomalies, thereby making explicit the circulation–surface link at seasonal time scales. Using ERA5 (1940–2024), seven year-round WRs and four seasonal WRs (DJF/JJA) are derived from Z500 over the Euro-Atlantic sector via EOF analysis and k-means clustering. A residual neural network takes as input monthly WR indices and calendar information, and reconstructs anomaly fields over Europe.

The model achieves high anomaly correlation and low error across large parts of Europe, especially in winter, and substantially outperforms classical linear WR-composite reconstructions. When the model is driven by the WR indices predicted by the bias-corrected SEAS5, it achieves comparable or better performance across most of the evaluated metrics. To address the question “How accurately do we have to predict the monthly mean WR indices to obtain a seasonal forecast of two-meter temperature and total precipitation that is better than SEAS5?”, we systematically degrade the WR indices, quantify how reconstruction skill depends on WR forecast accuracy, and identify the threshold beyond which the AI reconstruction surpasses the ECMWF SEAS5 seasonal forecast in reproducing European temperature and precipitation anomalies for the winter and summer seasons.

Results demonstrate that a large fraction of the spatial structure of European monthly anomalies can be inferred from the low-frequency Euro-Atlantic regime state. This provides a quantitative basis for AI approaches that exploit regime predictability to enhance sub-seasonal to seasonal forecast of European weather anomalies and related risks.