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

Despite recent advances, forecasting European weather on a seasonal timescale remains challenging for both numerical and statistical methods. Weather regimes (WRs), which represent recurrent, quasi-stationary, and persistent states of the atmospheric circulation, are well known to exert considerable influence over the European weather, offering a promising window of opportunity for sub-seasonal to seasonal forecasting. However, while much research has focused on the study of the correlation and the impacts of the WRs on the European weather, the estimation of ground-level climate variables, such as temperature and precipitation, from Euro-Atlantic WRs remains largely unexplored and limited to linear methods.

In this study, we present an AI model designed to reconstruct monthly mean anomalies of the European temperature and precipitation based on the dominant WRs. The model can capture and introduce complex non-linearities in the relation between multiple WRs, describing the state of the Euro-Atlantic atmospheric circulation, and the corresponding surface temperature and precipitation anomalies in Europe. The ability to reconstruct anomalies from WRs constitutes only a portion of the overall challenge. Predicting WRs on a monthly timescale is inherently difficult, and such forecasts are inevitably affected by errors, which can propagate and influence the quality of the reconstructed anomalies. In view of future developments, we examine the effect of inaccuracies in the WRs estimation on the anomalies reconstruction, establishing a lower bound on the WRs prediction accuracy required to outperform the ECMWF seasonal forecast system, SEAS5.

The model utilizes the monthly averages of weather regimes (WRs) to reconstruct the monthly averages of two-meter temperature and total precipitation anomalies during winter (DJF) and summer (JJA). ERA5 and NOAA-CIRES-DOE Twentieth Century Reanalysis datasets are used to compute the WRs and train the AI framework. Using ERA5 as the ground truth, the reconstruction performance is assessed through commonly used metrics, including mean squared error (MSE), anomaly correlation coefficient (ACC), and coefficient of efficiency (CE).

The results presented underline the importance of developing reliable WRs forecasting methods alongside reconstruction models to fully realize the potential of WRs-based forecasting systems. Our findings demonstrate that WRs-based anomaly reconstruction powered by AI-tools offers a viable pathway to better understand and predict seasonal variations.