Conditional Verification of ECMWF Medium-Range Forecasts

The predictive skill of the Integrated Forecasting System (IFS) from the European Centre for Medium-Range Weather Forecasts (ECMWF) has seen substantial improvement over recent years. Despite this progress, systematic errors persist, and their magnitude varies with forecast lead time, terrain complexity, and prevailing weather regimes. The assessment of conditional systematic errors as well as their relationship with predictive skill is crucial for understanding how these different conditions affect the forecast skill and uncertainty of near-surface variables. We evaluate deterministic and ensemble forecasts from ECMWF’s physics-based IFS and artificial intelligence-based AIFS systems, including the high-resolution IFS simulations at 4.4 km developed within the Destination Earth initiative. The results regard the latest update to the IFS Cycle 49r1, which among many other changes, includes the assimilation of 2-meter temperature observations, the activation of the Stochastically Perturbed Parametrizations (SPP) scheme for model uncertainty, as well as improvements of the land-surface modelling and assimilation methodology. Verification metrics—such as the root mean square error (RMSE), ensemble spread and bias—are analyzed across various synoptic conditions and orographic settings. Preliminary results indicate that forecast performance is strongly modulated by synoptic regime and topographic complexity. High-resolution forecasts show large improvements in near-surface fields over mountainous regions, while AI-based approaches can provide substantial skill gains even over flat terrain in certain conditions. These findings highlight the importance of accounting for atmospheric processes, model resolution, and orographic effects when evaluating different types of weather prediction models to better assess the forecast skill benefits and limitations and guide future model developments.