image: Figure 1: How the AIFS Single and Ensemble work. Copyright: ECMWF 2025.
Credit: ECMWF
Embargo Tuesday 1st July 2025 06:00AM BST
Today, just over 100 days after the launch into production of the world first openly available 24/7 operational AI forecast model AIFS-Single, ECMWF (European Centre for Medium-Range Weather Forecasts) is unveiling the first ensemble model using Artificial Intelligence/Machine Learning. The new model, called AIFS ENS, will be available as open source to the user community over the coming weeks.
The new ensemble model outperforms state-of-the-art physics-based models for many measures, including surface temperature, with gains of up to 20%. At the moment, it works at a lower resolution (31km) than the physics-based ensemble system which remains indispensable for high-resolution fields and coupled Earth-system processes; ECMWF is therefore also exploring hybrid systems that leverage the strengths of both approaches.
This high accuracy ensemble model complements the portfolio of ECMWF services by leveraging the opportunities made available by machine learning (ML) and artificial intelligence (AI). The AIFS ENS relies on physics-based data assimilation to generate the initial conditions, but it can generate forecasts over 10 times faster while reducing energy consumption by approximately 1,000 times compared to traditional ensemble forecasting methods.
In February, ECMWF launched the first operational data driven model called AIFS Single. This model runs a single forecast at a time, known as a deterministic forecast. Despite its accuracy, there is much more value to users if they can access the full range of possible scenarios. This is known as ensemble forecasting, a technique developed and implemented by ECMWF more than thirty years ago.
Dr Florence Rabier, Director-General at ECMWF, states: “ECMWF has now created an operational collection of 51 different forecasts with slight variations for our Artificial Intelligence Forecasting System (AIFS), which is a significant achievement and complements our physics-based products. But importantly, it’s not only us who are innovating. We are also working with and for 35 nations to advance weather science to improve global predictions. The availability of the AIFS ENS in conjunction with other ECMWF services will positively impact how national weather and meteorological services in our 35 Member and Co-operating States and beyond will be able to make their predictions and contribute to a safer society.”
ECMWF’s Director of Research, Dr Andy Brown, said: “This new milestone demonstrates our dedication to science-led innovations that are focused on delivering a machine learning forecasting model which pushes the boundaries of efficiency and accuracy, and it underscores our commitment to harnessing the power of machine learning for the weather forecasting community.”
ECMWF is leveraging the potential of what AI/ML can do for weather science with this latest model. This is part of its co-development of the award winning Anemoi framework with many of its Member States, which provides an open-source framework for training AI forecasting systems, including the AIFS.
ECMWF’s Director of Forecasts and Services, Florian Pappenberger, added: “We see the AIFS and IFS as complementary, and part of providing a range of products to our user community, who decide what best suits their needs. Making such a system operational means that it is openly available and comes with 24/7 support for our meteorological community. We will continue to engage with our Member States and our user community to ensure more and more parameters are added to suit their ongoing needs, and we will continue to enhance the model offered in line with how we push the capabilities of our physics-based system.”
For more information, members of the press are welcome to join the user webinar on Tuesday 24 June 10:00 BST. Please register here: https://events.ecmwf.int/event/487/
For interview requests, please contact ECMWF press office at pressoffice@ecmwf.int or Lorna Campbell 0044 (0)7836 625999.
Ends
Method of Research
News article
Subject of Research
Not applicable