Recent progress and outlook for the ECMWF Integrated Forecasting System

This presentation will provide an overview of recent scientific and technological developments at ECMWF, including the 2024 upgrade of the Integrated Forecasting System (IFS).

IFS Cycle 49r1 was implemented operationally on 12 November 2024 and substantially improved near surface wind and temperature predictions, especially in the Northern Hemisphere winter season. Key changes to the forecast model in Cycle 49r1 included the introduction of the Stochastically Perturbed Parametrisations (SPP) scheme for model uncertainty, improvements in wave and convective physics, and updates to land surface and atmospheric composition models.

Updates to data assimilation and observation usage in Cycle 49r1 included the assimilation of 2m temperature observations, assimilation of additional satellite data over sea ice, improved modelling of ocean emission and reflection and all-sky assimilation of AMSU-A, several changes to the land data assimilation system, and the introduction of version 13.2 of the radiative transfer model, RTTOV. The grid spacing of the Ensemble of Data Assimilations (EDA) was also reduced from 18 km to 9 km, with the inner loop minimisation grid reduced from 100 km to 40 km.

The impacts of Cycle 49r1 include substantial improvements to 2m temperature and 10m wind speed forecasts, increased spread for tropical cyclones intensity forecasts, a slight reduction of extreme wind forecast errors, and changes to representation of snow cover and snow density. The land data assimilation and model changes lead to a systematic reduction of soil moisture and higher spatial variability in soil moisture levels. At sub-seasonal lead times, Cycle 49r1 has small but statistically robust impacts on ensemble spread, which are driven by the switch to SPP scheme for model uncertainty. These changes are most evident in the tropics, where ensemble spread in the free atmosphere is reduced by several per cent, which represents a slight improvement in ensemble reliability relative to Cycle 48r1. In addition, Cycle 49r1 slightly improves the skill of Madden–Julian Oscillation (MJO) forecasts during weeks 3-4.

Cycle 49r1 is also the foundation for Cycle 49r2, a non-operational IFS cycle that will introduce new versions of the NEMO4/SI3 ocean and sea-ice model and underpin the 6^th generation atmosphere and ocean reanalyses (ERA6/OCEAN6), the new version of the atmospheric composition reanalysis (EAC5), and the next seasonal prediction system (SEAS6).  

In parallel to ongoing development of the IFS, ECMWF has developed the Anemoi machine learning toolbox to facilitate the development of data-driven weather prediction models, including deterministic and ensemble variants of the ECMWF AIFS. Real-time evaluation of pre-operational AIFS configurations has demonstrated that they are capable of very skilful medium-range forecasts for a range of upper-atmosphere variables, surface weather variables, and tropical cyclone tracks. The first operational version of the AIFS will be implemented later this year.

Finally, higher-resolution modelling capabilities are being accelerated by Digital Twin developments under the European Commission Destination Earth programme, which is building km-scale capability for a range of potential future HPC architectures. Major efforts are being invested in the code scalability of the Integrated Forecasting System to be able to run on GPUs and investigate alternative dynamical core options.