Artificial intelligence (AI) is profoundly reshaping weather forecasting, marking an era of enhanced accuracy, detail, and speed in predictions. The technology’s ability to analyze vast datasets is changing forecasts, ranging from immediate ‘nowcasting’ to sub-seasonal outlooks up to two months ahead. Such advancements promise significant benefits, including improved public safety, greater efficiency for weather-dependent industries, and vital tools for climate adaptation. The World Meteorological Organization (WMO) confirmed its “strategic shift to integrate Artificial Intelligence (AI) to advance Earth system science,” a move already decided by the World Meteorological Congress in 2023.
However, this AI-driven progress confronts serious challenges. The critical foundation of these models – open access to global weather data – faces jeopardy. The Financial Times reports that the Trump administration’s proposed 2026 budget sought to cut the US National Oceanic and Atmospheric Administration’s (NOAA) funding by $1.5 billion, or 24 percent. Since early 2025, over 550 employees have left NOAA’s National Weather Service (NWS).
Concerns have intensified, with five former NWS directors issuing an open letter warning that staffing shortages, saying “Our worst nightmare is that weather forecast offices will be so understaffed that there will be needless loss of life.” Rising geopolitical tensions also threaten the free flow of information. Furthermore, issues like the ‘black box’ nature of AI model interpretability and the need for ongoing validation persist. WMO Secretary-General Celeste Saulo noted that “we are far off track from achieving global climate goals,” highlighting the urgency.
Global AI Initiatives Transform Forecasting Capabilities
The European Centre for Medium-Range Weather Forecasts (ECMWF) is a key player. Its first operational AI model, launched in February 2025, improved accuracy by about 20 percent for crucial metrics like tropical cyclone paths, the Financial Times detailed. Florence Rabier, ECMWF director-general, explained to the Financial Times that advanced satellite data has largely closed the historical forecast accuracy gap between the Northern and Southern hemispheres.
The ECMWF also developed the Probability of Fire (PoF) AI model. This system forecasts wildfire activity by tracking vegetation and ignition sources, as detailed in a Nature Communications article, and has been operational since 2023 via the Copernicus Emergency Management Service. To further spur innovation, ECMWF, with WMO endorsement, launched the “AI Weather Quest”, a global competition to advance sub-seasonal forecasting.
Meanwhile, the China Meteorological Administration (CMA) is also embracing AI. It hosted a World Meteorological Centre Beijing Workshop in November 2024. JIN Ronghua, Director General of CMA’s National Meteorological Centre, explained that the workshop aimed to foster an exchange of ideas on AI in meteorology, discuss new technological opportunities and challenges, deliberate on WMO pilot projects, and support the UN’s Early Warnings for All initiative.
Tech Giants and Innovators Drive AI Weather Models
A new generation of ‘end-to-end’ AI systems is emerging. These models work directly from raw observations. Aardvark Weather, from the University of Cambridge, The Alan Turing Institute, and Microsoft Research, exemplifies this. It offers faster, lower-cost forecasting on standard desktop hardware, potentially democratizing access.
Scott Hosking of the Alan Turing Institute observed, “Suddenly we’re in this place where a new sensor can be set up and we can ingest that into the model very quickly.” Tech giants are significantly investing. Google DeepMind’s GenCast model, launched December 2024, reportedly surpassed traditional systems in 97.2% of scenarios.
Google also partnered with ECMWF on NeuralGCM and is expanding its MetNet nowcasting model to Africa. Microsoft’s Start weather team gained recognition for accuracy and developed models like Aurora. Nvidia’s CorrDiff model aims for ultra-high-resolution local forecasting. Dion Harris from Nvidia explained that AI allows for thousands of ensemble members, improving extreme event prediction.
NASA and IBM also contributed with the Prithvi WxC open-source AI model. Specialist start-ups like Brightband, Silurian, and Tomorrow.io are also innovating. Shimon Elkabetz, Tomorrow.io’s CEO, told the Financial Times, “When we started everyone said it would be too expensive to build our own constellation, but the new space economy is enabling us to do things that weren’t possible before.”, regarding their satellite constellation.
Navigating Data Dependencies and Model Limitations
The success of these AI models critically depends on continuous, open data. Richard Turner, a Cambridge University machine learning professor, emphasized to the Financial Times the fantastic level of international data-sharing. However, the NOAA crisis looms large.
Beyond the initial budget cut reports, PBS NewsHour revealed NOAA will no longer update its “Billion-Dollar Weather and Climate Disasters” database beyond 2024. Professor Turner expressed to the Financial Times that the potential data loss “is a huge worry. The community hasn’t — surprisingly, in my view — woken up to this danger yet… Yes, there is massive concern on this and I think the cuts are very dangerous at a time when the climate really is changing.”
Despite AI’s power, limitations remain. Traditional models may still outperform AI in predicting specific aspects like storm intensity. The so-called ‘black box’ issue also complicates understanding AI-driven outputs. Peter Neilley of The Weather Company remarked to the Financial Times on how “How rapidly this has overtaken at least the weather forecasting part of our science is truly remarkable. It really occurred in the last five years and it’s accelerating.”
However, Florian Pappenberger, ECMWF deputy director, says that AI models learn underlying physics, not just past patterns, enabling them to predict unusual events.
The Evolving Role of Meteorologists in an AI Era
AI is transforming, not replacing, human meteorologists. Kirstine Dale, Met Office chief AI officer, told the Financial Times, “We see the potential for a real step change… in how we forecast, which is in some ways similar to when we started using computers”, and foresees “an increasingly symbiotic relationship. We need them to work together on a team.”
Human expertise will remain vital for adjudicating between models, contextualizing forecasts, and communicating risks. The UK Met Office is exploring hyperlocal, street-level temperature predictions, which Ms. Dale described to the Financial Times as “That’s like street-level forecasting.”
Robert Lee, a Reading University meteorologist, explained that while precise day-by-day forecasting is limited beyond two weeks, AI can predict general conditions for longer periods, offering significant benefits, especially to the energy sector by anticipating extended cold or stormy periods.
