Researchers from FZJ, DWD, ECMWF, and the University of Bonn convened at DWD’s headquarters in Offenbach for the first progress meeting of the RAINA project. Discussions focused on integrating RAINA with the WeatherGenerator model, whose prototype has recently made available on GitHub. Key topics included dataset selection for training, with ICON-DREAM and RADKLIM identified as strong candidates for improving high-resolution forecasts of extreme wind and precipitation events. ICON-DREAM provides a multi-year depiction of the global atmospheric state, featuring a refined nest over Europe, while RADKLIM offers over two decades of rain gauge-adjusted precipitation observations at kilometer scale from the German radar network.
A recent effort is the development of a fully machine learning-based system that fuses GraphCast forecasts, originally developed by Google DeepMind, with statistical downscaling to generate precipitation and wind gusts predictions over Germany at approximately 2 km resolution. DWD’s verification suite supports the evaluation of these forecasts, with a special focus on refining spatial object verification for improved precipitation modelling.
The meeting also fostered collaboration with KI-HopE-DE, a sister project from BMBF’s funding measure for the development of flexible, resilient and efficient machine learning models. KI-HopE-DE aims to improve flood forecasting in small and medium-sized river basins. Discussions highlighted the potential of RAINA’s improved precipitation forecasts in enhancing flood prediction models and emphasised the shared interest in probabilistic forecasting techniques.
The event concluded with productive discussions and a strong outlook for future advancements in high-resolution weather forecasting.
