50 years of weather forecasting at the ECMWF

I have a PhD in meteorology and have spent my entire career in the field of Numerical Weather Prediction, moving between Météo-France, the French Meteorological Service, and ECMWF, in Reading, in the United Kingdom. My interest in the field developed from my passion for science and from my childhood experience. I grew up in the southwest of France, close to the Atlantic Ocean and the Pyrenees mountains, where I could follow the changes in the weather. I was fascinated by the alternations—one day would be cold and rainy with northwesterly winds, the next clear and warm with a southerly breeze. I also witnessed major storms and huge waves. This gave me an early sense of the challenges, but also the usefulness of producing accurate weather predictions.

Aside from the intellectual satisfaction derived from working on a state-of-the-art system, what provides a real motivation is knowing that the science we develop at ECMWF has a clear purpose: delivering useful data for society. ECMWF forecast data are provided to our Member States’ national meteorological services and are used, together with their own forecasts, to help protect life and property. Knowing where and when a storm will hit or how intense a heatwave will be days and weeks in advance is of prime importance for society to be prepared. Our data are also distributed more broadly to other meteorological services around the world and used to provide services for the benefit of national economies. Many sectors are highly dependent on meteorology: agriculture, transport, construction, tourism, and energy, particularly renewable energy, to name a few. Also, documenting climate change, air pollution, floods, and fire risks, in the context of our partnership with the European Union for the Copernicus Programme, is highly motivating. Every improvement made in our output has a direct translation in terms of societal benefit. This provides extra motivation to do one’s best when you know that so much is at stake.

ECMWF is an intergovernmental organisation supported by 35 countries across Europe, which together provide the strategic direction of our work. We are widely recognised as a world leader in numerical weather prediction, providing high-quality data for weather forecasts and environmental monitoring. So first, what are medium-range weather forecasts? They are predictions that estimate weather conditions for a period that typically covers about 3–10 days into the future. These forecasts bridge the gap between short-range (up to 3 days) and long-range (beyond 10 days) prediction, providing useful guidance for planning and preparedness over the coming week and beyond. How do we achieve such accurate weather forecasts? We need three ingredients: a numerical model, a computer, and observations. Our numerical model is a huge computing program with millions of lines of code, which encompasses all of our knowledge of the physical laws that govern the atmosphere and its interactions with the ocean, the land, and sea ice. This model is performing its calculations several times a day on a supercomputer (currently with a processing power capable of performing 30 million billion calculations per second). Results of all these computations are sent to all of our users and are also stored in one of the largest meteorological data archives in the world maintained by ECMWF, to be used for research and future developments. The forecasts generated by this model are re-adjusted at regular intervals with observations coming from all over the world, from weather stations, ocean buoys and satellites. Satellite data account for more than 95% of the total of all observations used at ECMWF, and we maintain strong partnerships with European space agencies to make the most of this fantastic resource.

Our vision is that world-leading monitoring and predictions of the Earth System enabled by cutting-edge physical, computational and data science will contribute to a safe and thriving society.

ECMWF was established 50 years ago by a group of European nations who had the vision of a collective endeavour to improve weather forecasting. At the time, weather predictions were reliable up to around 1 to 2 days ahead, and anything further than a few hours was notably unreliable. Being able to predict the weather accurately and reliably for longer ranges had become imperative. And that decision marked the birth of ECMWF on 1 November 1975, through the signing of a Treaty.

What once seemed almost impossible in the ‘70s has become reality: we are now providing useful forecasts, not only in the medium range but also up to a season or even a year ahead!

Although the evolution of the forecasting system has been gradual in the last five decades, there have been some notable breakthroughs, sometimes disruptive, which have enhanced the breadth and reach of our activities. A few of these are highlighted below:

• Ensembles: Realising that the value of forecasts is much larger if there is a proper estimate of their uncertainties, ensemble forecasts have become the method of choice. This means that 50 parallel forecast computations are performed in parallel to document the range of possible futures.

• Use of satellite observations: Building on the theory of optimal control, a new method to blend models and time-distributed observations was implemented at the end of the ‘90s. Combined with the advent of advanced satellite data, this method led to a large improvement in forecasts.

• Reanalysis: Going back in time to re-process data and reconstitute accurate weather conditions from decades ago to now is essential for documenting climate change. Today, with the advent of machine learning (ML) in weather prediction, this so-called “reanalysis” provides the basis for training ML models.

• Earth System: Weather prediction is no longer based solely on an understanding of the atmosphere. Today, our model runs a full Earth system for weather, which does not only include the atmosphere, but also the land, ocean, sea ice, atmospheric composition, and their interactions.

• Longer time ranges: As model quality improved and became coupled with other components of the Earth System, such as the ocean, ECMWF has gradually extended its forecasts to longer time-ranges, going from medium to seasonal timescales.

• Hybrid computing architectures: Starting with our first supercomputer, the well-known Cray 1, installed in our headquarters in Reading In 1978 (the first Cray in Europe!), supercomputers have steadily increased in power and are now adopting hybrid architectures, with a mix of CPUs and GPUs. This required us to completely revisit how we write our numerical codes, making them more portable and flexible.

• Machine Learning (ML): Fully embraced by the organisation, ML is bringing invaluable benefits to our area of work. ML models, once trained, can perform faster and cheaper than physics-based models. We are also very well placed to train such models as we produce the data commonly used for such purposes.

Since the birth of ECMWF, collaboration with partners, primarily with our Member States and partners in Europe, including satellite agencies like the European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT) and the European Space Agency (ESA), has been key. A key partnership that has grown over the years is our collaboration with the European Union, with which we operate Copernicus services and projects in the field of climate change, air quality, but also groundbreaking work on the concept of Digital Twins of the Earth in the context of the Destination Earth initiative. These initiatives are absolutely crucial in the context of anthropogenic climate change, as described below.

The Copernicus Climate Change Service provides quality-controlled information about the past, present and future of the climate in Europe and worldwide. Data and tools are provided for use by governments, public authorities, policy makers and private entities across multiple disciplines. Among the most popular datasets are the so-called reanalyses, which describe the state of the atmosphere, land surface and ocean waves, providing the best possible picture of past weather and climate conditions, as previously explained. Reanalyses form the basis for monthly climate bulletins and are used in the assessment of the State of the Climate in Europe, published annually by the Copernicus Climate Change Service and the World Meteorological Organization (WMO). They are also regularly cited and used in the Intergovernmental Panel on Climate Change (IPCC) Assessment Reports to underpin their findings and support informed recommendations.

In parallel, the Copernicus Atmosphere Monitoring Service is Europe’s leading service for monitoring air quality, climate patterns, and UV radiation levels. It provides twice-daily information on global and regional air quality, inventory-based emissions, observation-based surface fluxes of greenhouse gases and from biomass burning, solar energy, ozone and UV radiation. Tens of thousands of users access its platform directly, while over 200 million end users worldwide benefit via smartphone applications, websites, or TV bulletins. The introduction of new components, such as monitoring anthropogenic CO2 emissions, demonstrates the service’s evolution towards providing comprehensive environmental data for informed decision-making at a global scale.

The Copernicus Emergency Management Service, led by the Joint Research Centre (JRC), provides early warnings of natural disasters. ECMWF contributes to the European Flood Awareness System, the Global Flood Awareness System and the European Forest Fire Information System.

Finally, Destination Earth (DestinE) is an initiative to digitally enable European environmental programmes. It aims to develop a highly accurate digital replica of the Earth, called a digital twin, to model and simulate natural phenomena, hazards, and the related human activities influencing them. The initiative is implemented by ECMWF, ESA and EUMETSAT, with the participation of many European organisations as contractors. DestinE attempts to unlock the potential of digital modelling of the Earth system at a level that represents a real breakthrough in terms of accuracy, local detail, access-to-information speed and interactivity. By combining an unprecedented amount of data with innovative Earth system models and cutting-edge computing, DestinE allows users to explore interactively the different components of the Earth system (land, marine, atmosphere, biosphere) and the natural and human-induced change, looking at the past and present while testing and developing future scenarios. ECMWF is responsible for implementing two high-priority digital twins (on climate adaptation and weather extremes) and the Digital Twin Engine, the software and data environment needed to power the digital twins. The initiative is also investing in dedicated technology partnerships with High-Performance Computing (HPC) centres, vendors and international organisations representing the entire technology range across extreme computing, big data, smart sensors and networks.

Meteorology is quite unique in its collaborative and cooperative nature. Beyond collaboration in Europe, ECMWF is very engaged with the WMO, contributing to many committees, expert teams and activities. We also have strong links with agencies around the world in the United States, China, Africa, Brazil, Japan, Australia and Canada. We all face the same problems and benefit from exchanging experience, data, and occasionally staff. Although these formal cooperation agreements are not essential for scientific interaction, they frame the collaboration and allow the partners to focus on areas of common interest.

Artificial Intelligence (AI) has certainly taken the world by storm, and meteorology is no exception. This is undeniably the most recent challenge (and opportunity) faced by weather prediction, and also the most exciting. The possibilities provided by machine learning in our field are immense and the first results show tremendous potential. Not only are the results outstanding, but the computational power necessary to achieve predictions is greatly reduced. This is too significant of an opportunity to miss! The priority for ECMWF has been to embrace and exploit machine learning, in collaboration with Member States and the scientific community, while keeping our system anchored in the physics-based approach which has driven our success in the last few decades.

A newly operational model, known as the Artificial Intelligence Forecasting System (AIFS), was launched by ECMWF this year. For many measures, including tropical cyclone tracks, the AIFS outperforms state-of-the-art physics-based models, with gains of up to 20%. This highly accurate model complements the portfolio of ECMWF’s physics-based models, advancing numerical weather prediction and leveraging the opportunities made available by machine learning and AI, such as increased speed and a reduction of approximately 1000 times in energy use per forecast.

Among the available AI models, the AIFS provides the greatest granularity sought by its user community. On top of vital fields for users, like wind and temperature, and details on precipitation types from snow to rain, this new service is the first fully operational weather prediction open model using machine learning with the widest range of parameters. The AIFS has been designed holistically with all users in mind. For example, in the renewable energy sector, it will help with predictions, such as surface solar radiation levels or wind speeds at turbine levels, so that operations can be maximised.

In 2025, it is clear that the needs of society for accurate weather and climate monitoring information have never been greater. Reliable advance warnings of weather events are essential for the protection of lives and properties, and environmental monitoring is a crucial input for developing appropriate adaptation and mitigation strategies.

A great opportunity in this context is the shift towards a convergence of AI and physics-based meteorology, as explained above. This shift towards AI/ML has been catalysed by the availability of high-quality, openly available reanalysis datasets, advances in deep learning architectures, and the demonstration that AI models can generate global forecasts at a fraction of the computational cost of traditional methods. ECMWF has played a key role in this transformation, developing operational ML-based forecasting systems, as explained above, and fostering international collaboration through open-source frameworks.

The careful adoption of machine learning in weather prediction stands to benefit both science and society, extending the boundaries of what is possible in forecasting. But with this opportunity come challenges.

One challenge is the fostering of collaboration between public and private actors, including Big Tech, that have shown a keen interest in this field, and are new players in meteorology. It is also essential to narrow the digital divide with countries in the Global South, ensuring that no one is left behind. In this context, the challenge for the global community is to invest in shared, open data and AI tools that are trusted, transparent, interoperable, and tailored to the needs of diverse users.

Meteorology and climate sciences are fields of the future, and there has never been a better time to begin a career in these areas. What makes meteorology unique is the openness that exists: observations are freely exchanged all over the world and the weather knows no national boundaries.

Advanced research in mathematics, physics and data science can both be directly used by and influence society. There are few other subject areas that can offer such a combination of excitement, challenge and tangible impact.

My hope for the future is that, by joining forces between governments, academia and the private sector, we will find solutions to the climate crisis and that, through technology and the widespread availability of data and tools, all countries will soon have easy access to early warnings of extreme events and be better prepared for what lies ahead.