Recent Developments in R2O at ECMWF

The European Centre for Medium-Range Weather Forecasts (ECMWF) has developed a comprehensive R2O process over the past five decades which is used to implement new versions of the Integrated Forecasting System (IFS) into operations. This process is now also being applied to global reanalyses of meteorology and atmospheric composition. An ever-growing number of users with a diversity of requirements, especially for the timely availability of increasingly comprehensive test datasets, fuels the design and implementation of new meteorological data-generating initiatives.

The adoption of such new initiatives requires the R2O process to be broadened and updated to ensure a robust transition from research/development to operations for these initiatives. A brief description of the R2O process applied at ECMWF will be given in the context of managing operational forecast model upgrades, and in producing new global reanalyses products, such as the next ECMWF global meteorological reanalysis (ERA6) and the upcoming atmospheric composition reanalysis (EAC5), which will be generated under the European Union-funded activities by the Copernicus Climate Change (C3S) and Atmosphere Monitoring (CAMS) Services, respectively.

Exciting new initiatives, such as the European Union-funded Destination Earth programme, where high-resolution (km-scale) numerical weather predictions are being run on HPCs external to ECMWF (EuroHPC), and the rapid adoption of data-driven, machine-learned models applied to numerical weather prediction, present interesting challenges and opportunities to further evolve the R2O process. The subsequent integration of the transition from research to operations for these initiatives into ECMWF’s R2O process will be presented. Future aspects which may need to be considered within R2O will be explored, such as increasing use of data-driven forecasts and their associated training requirements, handling a blend of physical and data-driven forecast models, potential demand for the ability to run forecast models on external HPC or cloud infrastructures, and the desire to perform more continuous deployment.