In our latest podcast episode, experts from leading meteorological institutions discuss how Artificial Intelligence is impacting the overall organisation of services and activities, highlighting the crucial role of European collaborations in driving this technological shift. Joining the conversation are Virginie Schwarz, General-Director of Météo-France; Roland Potthast, Head of Numerical Weather Prediction at the German Weather Service (DWD); and Florian Pappenberger, Deputy Director General and Director of Forecasts and Services at ECMWF.
With the rise of Artificial Intelligence (AI) in recent years, Europe’s weather prediction landscape is undergoing a revolution. AI is not only enhancing the precision and the speed of forecasts, but it is also reshaping the roles and responsibilities of key institutions at national and European level. In this context, the longstanding European collaborations around Numerical Weather Prediction, supported by initiatives like Eumetnet or Destination Earth, provide a crucial framework to manage the complex developments that come with integrating AI into weather prediction. In this episode, our experts discuss the strategic impacts of AI-driven transformations and their broader implications for weather and climate predictions.
AI for weather forecasts, what changes?
The discussion began by identifying the fundamental aspects driving this transformation, helping the audience better understand the key innovations.
“Machine learning has come as a storm, demonstrating a quality of AI-based forecasts which we didn’t anticipate, and really starting to either equal or outperform our traditional modeling approaches” said ECMWF’s Florian Pappenberger.
This animation shows a 10-day forecast of meridional wind at 850 hPa using ECMWF’s Artificial Intelligence Forecasting System (AIFS).
Unlike traditional weather forecasting which relies on physical models and the immense power of supercomputers, AI leverages data-driven approaches allowing for new possibilities such as faster forecasts with fewer computing resources, leading to changes across the entire forecasting value chain. This is primarily seen as a philosophical change from the perspective of the French meteorological service, Météo France’s General Director, Viriginie Schwarz: “With these inductive models, we learn from past data but without having this physical description of all the underlying mechanisms. This is a radical change in our way of thinking and working, both for the researchers and the forecasters. And we are probably going to keep-on discovering things going against our practice and our intuition.”
This position was shared by Roland Potthast of DWD:
The verification and the evaluation of the system are changing. This brings questions to the classical modellers on how to change their thinking and use their knowledge for further AI developments. There is a whole science of bringing classical ideas into the AI framework, and that is just starting. We are at the beginning of a revolution and new evolution.
Roland Potthast, DWD
AI-based weather forecasting models create new opportunities to complement traditional methods. One area is ensemble forecasting, the practice of running multiple forecasts to estimate uncertainty, which is typically computationally intensive. And this has also broader implications.
“This leads us to completely rethink, where, when, and how weather services are produced, and opens the possibility to do things such as having much more members in an ensemble, things that were not accessible previously” said Virginie Schwarz.
The impacts on roles and responsibilities of the European meteorological infrastructure
Moving forward, the discussion also looked at how the European meteorological infrastructure (EMI) is adapting to these changes, particularly in reshaping meteorologists’ roles to exploit AI.
Considering these evolving dynamics, a common viewpoint emerged.
At the core, our fundamental objective remains the same: providing to public authorities, companies, and the public the best information and decision-support services, based on high scientific and human expertise. But what will change is how we do that, what we deliver.
Virginie Schwarz, Météo-France
Florian Pappenberger fully supported this view from ECMWF’s perspective as well, adding that: “At its core, our mission also remains the same, providing the best forecasts and data to support our Member States doing the job they are mandated to do in the best possible way. But the soul searching starts regarding the “how”, considering now this new dual approach.” This covers elements like resource allocation, from computing power to staffing, and the types of observations to be collected.
The speakers also discussed the “democratisation of forecasts”, a transformative possibility enabled by AI advancements that could allow people to run sophisticated forecasting models directly on their laptops. Florian noted that such a shift would also prompt organisations like ECMWF to reassess how they could further support their Member States.
From the DWD perspective, Roland Potthast said: “Now that AI becomes a conversation partner to infer information in just one click, you want to know what is reliable. And I think this even strengthens our roles. We know how to judge information, how to process it, and how to work with it.“
Delivering the best possible weather services: the importance of the human element
Virginie Schwarz stressed the importance of the human element for delivering the best possible weather services by the mandated national institutions like Météo-France and DWD: “As a weather service, you are not just sending-out a forecast on the Internet. You are explaining the forecast, the uncertainties, the risks, the potential for change… And I think this human element, when we are talking about events that can have such huge consequences on the life of our fellow citizens, is also very important and will remain key in the future.”
Challenges
AI-based forecasting models require intensive training phases, involving significant computing resources, before they can be effectively deployed as forecasting tools. Accelerator technology in next-generation supercomputers is crucial in this regard, making ongoing developments in this area essential for expanding AI applications across various fields.
Also crucial are high-quality datasets used to train the AI models, such as the ERA5 reanalysis produced by ECMWF in the framework of the Copernicus Climate Change Service, or regional predictions produced by the national meteorological services.
Roland Potthast highlighted that “This is also a massive migration challenge of know-how and people on how to work with these systems. “
Virginie Schwarz pointed out that the allocation of resources is also a challenge: “Because we are dealing with these two systems, we have to really find the means to have both the human resources to train the people, and the computing resources, the GPUs, that we need for the new developments.”
She added: “We really to stay in the game and try to manage the transition as best as we can, keeping in mind this objective that we are here to serve the public. We are here to help manage the climate transition and adapt to the effects of climate change, and this doesn’t change.“
Concluding the topic, Florian Pappenberger noted that some challenges can also present opportunities:
The National Weather Services have been at the forefront of delivering effective warnings for many years. When we integrate this huge body of knowledge and expertise with the novel possibilities brought by machine learning, the European Meteorological Infrastructure can really augment weather prediction services in a way in which nobody else can.
Florian Pappenberger, ECMWF
The strengths of collaborations around AI to support adaptation and boost innovation
Building on existing strengths
Meteorological institutions across Europe have a longstanding tradition of knowledge-sharing and collaboration. With diverse expertise and know-how spread across both national and European institutions like ECMWF, this established network has supported innovation in climate and weather science for years. Now, with the AI revolution reaching every institute, our speakers have also shared their views on how these partnerships serve as a key asset in this transition.
“European collaborations are crucial for all of us. We have all been working in and beyond the numerical weather prediction consortiums framework for a long time. There’s a strong tradition and now it’s going into the next phase” said Roland Potthast.
As an important player of many joint initiatives, Météo-France Director Virginie Schwarz explained that: “Collaborations on Numerical Weather Prediction have been something that has structured Météo- France for decades. Both the joint development of our IFS ARPEGE code with ECMWF, but also the collaboration through the ACCORD consortium, which brings 26 countries together. This has been both something that Météo-France has strongly contributed to, but also benefited from. And this is really one of the things that have made our NWP systems what they are today.”
Florian Pappenberger also highlighted how this collaborative framework can serve as a catalyst to connect AI innovations with societal benefits: “I think these collaborations will get stronger, more integrated. But I also believe national weather services will be required to take up this machine learning revolution, and bring it to the end, to the users, ensuring that action is taken for the good and the safety of society.”
Growing AI partnerships
ECMWF has been at the forefront of global forecasting for many years, and now is picking-up AI as a joint enterprise with the weather services of its Member States.
Roland Potthast
With AI introducing new knowledge and technology into the mix, Virginie also explained how the nature of collaborations is also transforming: “This AI partnership is really the first time where, at this level, we are all going be working together, with all the modelling consortia, to try to bring forward some progress on meteorology and climate. “
Roland Potthast highlighted the value of Eumetnet activities, the collaborative framework on weather established between the national services, ECMWF and EUMETSAT, also looking at undertaking AI activities jointly.
Boosting the AI innovation with Destination Earth
What started as a traditional NWP collaboration project, is now becoming more and more an AI collaboration project.
Virginie Schwarz
Reinforcing this collaborative landscape, European projects have also been essential in this respect. A prime example is the Destination Earth initiative of the European Commission, uniting more than 100 partners across 25 countries. Its second phase includes important AI developments, aiming to help quantify uncertainty and enhance the interactive features of the Digital Twins developed by ECMWF and its partner organisations.
Virginie Schwarz said: “Météo-France coordinates the on-demand component of the Weather Induced Extremes Digital Twin. And it is very interesting to see that what started as a traditional NWP collaboration project, is now becoming more and more an AI collaboration project. And with the changes that we have seen coming with AI, it has been great that the European Commission has allowed us to bring this into the project.”
Anemoi: a successful collaboration
Cross-European collaborations on AI have produced impactful tools like Anemoi, a new framework for developing machine learning weather forecasts, helping to explore deployment and developments of AI models on the new generation of supercomputers.
Illustration of the promising results for a 7-day forecast of 10 m wind speed (shading) and sea-level pressure (contours), obtained with the regionally high-resolution AI-based model Bris, developed by MET Norway and partners, making use of the Anemoi framework. The model has learned to forecast at high resolution (here about 2.5 km) inside the Nordic region, and at low resolution (here about 30 km) outside of this domain. The model successfully creates a higher-resolution structure over the Nordics.
Roland Potthast said: “ECMWF ran into the problem first, and they came up with a solution, benefitting all of us as we are now using it. That is an example where this collaboration really pays-off.”
A view that was also endorsed by Virginie Schwarz: “Examples like the Anemoi framework that many of us will be able to use, really help us to be more efficient and to do things faster than on our own.”
The lively discussion on the topic of collaborations showcased how the combination of AI tools and cross-border expertise can unlock new possibilities, setting new benchmarks for what can be achieved in meteorological science.
Exploiting AI to enhance Europe’s resilience
The conversation also dwelt on how artificial intelligence can contribute to Europe’s efforts in building resilience against the effects of climate change and extreme events, by translating leaps in AI technology into societal benefits.
Virginie Schwarz said: “Increasing the geographical precision and the anticipation, especially of our extreme weather forecasts, is the key thing for us. This is how we can get better information to the authorities and the public, so that appropriate measures can be taken.”
Florian Pappenberger also highlighted how decision-making can be supported with the new capabilities: “Machine learning will allow us to have larger ensembles and quantify better where certain events occur and how likely they are. And we haven’t really touched on one of the big things which are emerging, which is climate predictions and sub-seasonal forecasting.”
Machine learning techniques also show great potential for supporting adaptation strategies around floods.
Roland Potthast said that: “It is not just the quality which improves, but AI systems can identify phenomena in the forecast, and they can communicate their findings downstream to the products. Then you see that we are in the middle of a real revolution wider than the forecast itself. “
The conversation concluded on outlooks into the future, inviting our participants to imagine how developments around AI will continue.
Florian Pappenberger said: “I think we are going to see a forecast directly coming from observations soon, and that we will need more investment into GPUs or similar technologies.”
Roland Potthast shared his vision, saying: “With now AI information processing, it will change the way we interact with data. The whole way down to the end user will change. It’s more unclear how it will spread in society. The future is open.”
Virginie Schwarz added that: “AI is going to turn us into real high-level experts in processing, storing and accessing data.” And concluded with her vision for the future: “It will allow us to adopt much more flexible ways of working. And really serve our customers better, in a more agile way, in a world where everyone expects to be updated very quickly.”
Destination Earth is a European Union funded initiative launched in 2022, with the aim to build a digital replica of the Earth system by 2030. The initiative is being jointly implemented by three entrusted entities: the European Centre for Medium-Range Weather Forecasts (ECMWF) responsible for the creation of the first two ‘digital twins’ and the ‘Digital Twin Engine’, the European Space Agency (ESA) responsible for building the ‘Core Service Platform’, and the European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT), responsible for the creation of the ‘Data Lake’.
We acknowledge the EuroHPC Joint Undertaking for awarding this project strategic access to the EuroHPC supercomputers LUMI, hosted by CSC (Finland) and the LUMI consortium, Marenostrum5, hosted by BSC (Spain) Leonardo, hosted by Cineca (Italy) and MeluXina, hosted by LuxProvide (Luxembourg) through a EuroHPC Special Access call.
More information about Destination Earth is on the Destination Earth website and the EU Commission website.
For more information about ECMWF’s role visit ecmwf.int/DestinE
For any questions related to the role of ECMWF in Destination Earth, please use the following email links: