Every five years ECMWF develops a new Strategy for the next ten years. This sets out anticipated progress and direction for the whole of ECMWF, from the science we develop and the impact we have to our organisation and people.
The latest Strategy was approved by our Council in December 2024 and covers the period 2025–2034. Here, we set out the main points this Strategy makes, including the emphasis on collaboration and on the growing role of machine learning and artificial intelligence in weather prediction.
The importance of collaboration
The Strategy points out that we work extremely closely with the national meteorological services (NMSs) of our Member and Co-operating States and with partners to deliver our mission.
The global predictions we produce complement the capabilities of the NMSs, and thus they help them to fulfil their roles and provide a better service for European society.
This is expressed in ECMWF’s Vision, which says that our “cutting-edge physical, computational and data science, resulting from a close collaboration between ECMWF and the Members of the European Meteorological Infrastructure, will contribute to a safe and thriving society”.
Machine learning in the new Strategy
There was already a role for machine learning (ML) and artificial intelligence (AI) in the previous Strategy: they were to be combined with physics-based conventional approaches in the establishment of initial conditions for weather forecasts (data assimilation) and in modelling.
This continues to be central, but progress in ML/AI has been very rapid in the last couple of years. This means that, as a priority, we shall make the Artificial Intelligence Forecasting System (AIFS) we have developed operational.
The AIFS is a forecasting system based on ML/AI, trained on many years of reanalysis and using traditional data assimilation. Our medium-range operational output will include a single ‘deterministic’ version of the AIFS and an ensemble version.
However, the Strategy goes beyond this. “We will also investigate the potential of ML models for other components of the Earth system,” says ECMWF Director-General Florence Rabier. “Moreover, there will be more research into using ML techniques directly on observations, to produce forecasts without using reanalysis or data assimilation.”
For 2035, the Strategy envisages ECMWF exploiting data-driven methods anchored on physics-based modelling.
This is an example of forecasts using ECMWF’s experimental Artificial Intelligence–Direct Observation Prediction (AI–DOP) system. They show forecasting day five (20 June 2022, 12:00 UTC) for (a) 2 m temperature, (b) temperature at 850 hPa and mean sea-level pressure, (c) sea-surface temperature, and (d) 10 m eastward component of wind. From an ECMWF Newsletter article.
Other developments in science and technology
The Strategy foresees increasing efforts to integrate ML approaches into data assimilation. It also anticipates that the resolution of 4D-Var data assimilation will increase, and the windows over which it takes place will become more continuous.
As well as better exploiting existing data, ECMWF will strive with its partners (such as Member and Co-operating States, EUMETSAT and ESA) to gain maximum benefits from new observations.
The Strategy emphasises a hybrid approach, with increasing levels of ML, to the modelling of the Earth system.
“Data assimilation and observations will be used to directly improve forecasting models,” says Director of Research Andy Brown. “Uncertain model parameters will be learnt, and ML will be used to apply a correction term in the model.”
This is an early example of using ML in data assimilation to achieve a time-dependent tendency correction. It shows a time series of constant tendency correction and time-dependent tendency correction of near-surface temperature between 00 UTC on 20 July 2022 and 10 UTC on 24 July 2022. (Time series courtesy of Patrick Laloyaux, ECMWF – see this news article for more details.)
We shall continue to work with partners in our Member States to drive down errors in the physical model and improve the representation of model uncertainty. There will be a particular focus on improving the skill of sub-seasonal predictions.
There will also be efforts to go down to kilometre-scale global modelling, taking advantage of our role in the EU’s Destination Earth initiative. For the long run, we will continue to develop a new nonhydrostatic core.
We want to continue to use a resilient, efficient, and cost-effective high-performance computing facility. Changing requirements include, for example, a greater emphasis on graphics processing units (GPUs), and there is a new focus on cybersecurity.
The impact of ECMWF’s activities
The Strategy states that we will continue to provide the best quality medium-range weather predictions to our Member States. This will be done with a particular focus on extremes and high-impact weather.
ECMWF will also produce datasets to support the training of ML models by ECMWF and its Member and Co-operating States.
Through our role in the EU’s Copernicus services, we will continue to serve user needs for wider environmental applications. These include climate monitoring, air quality, floods and fire, and support for climate prediction.
In addition, the Copernicus Atmosphere Monitoring Service (CAMS) is developing the CO2 Monitoring and Verification Support capacity (CO2MVS) to provide new information on anthropogenic greenhouse gas emissions.
We will proceed on our path towards open data and open-source software, and training and support for our Member and Co-operating States and users will remain crucial.
“A focus of the Strategy is the collaboration with our Member and Co-operating States as well as globally, for example with the World Meteorological Organization (WMO) and space agencies,” says ECMWF’s Director of Forecasts and Services, Florian Pappenberger.
Organisation and people
This section of the Strategy states that funding should be aligned with strategic objectives. We will adapt to the impact of ML/AI on opportunities, staff profiles, and training requirements, and initiatives related to Diversity, Equality, and Inclusion (DEI) should be advanced.
We will also continue along the path of net-zero carbon emissions, and we will ensure a smooth transition from our current facilities in Reading (in our host country, UK) and in Bonn (Germany) to planned new buildings.
The full Strategy
The full Strategy 2025–2034 and a summary version are available on our website.