About
ECMWF AIFS is the European Centre for Medium-Range Weather Forecasts' flagship machine learning weather forecasting system, representing a major leap forward in AI-driven meteorology. Built on the Anemoi framework—a collaborative open-source project involving ECMWF and national meteorological services across Europe—AIFS delivers global numerical weather predictions at a horizontal resolution of 28 km (0.25°). Unlike traditional physics-based models, AIFS leverages deep learning to produce fast, accurate forecasts for a wide range of variables including precipitation, 2-metre temperature, ocean wind waves, snow cover, sea ice, and tropical cyclone tracks. The system supports both deterministic (single-run) and ensemble (probabilistic) forecast modes, providing richer uncertainty estimates for high-impact events like severe storms. ECMWF has made the first AIFS model weights openly available, along with a training-ready version of the ERA5 reanalysis dataset, enabling researchers and developers worldwide to run, fine-tune, and build on state-of-the-art AI weather models. AIFS open data can be accessed directly from ECMWF's open data portal, opening new possibilities for research, education, and application development. The Anemoi framework underpinning AIFS provides shared tooling for data loading, model training, and inference, making it easier for meteorological organisations to adopt operational AI forecasting. AIFS is aimed at meteorologists, climate researchers, weather application developers, and data scientists seeking cutting-edge AI-based Earth system modelling capabilities.
Key Features
- ML-Based Global Weather Forecasts: Produces deterministic and ensemble global weather predictions at 28 km (0.25°) resolution using deep learning instead of traditional physics-based methods.
- Open Model Weights & ERA5 Dataset: AIFS model weights and a training-ready version of the ERA5 reanalysis dataset are openly available, enabling the community to run, adapt, and build upon state-of-the-art AI weather models.
- Anemoi Framework Integration: Built on Anemoi, a collaborative open-source ML framework co-developed by ECMWF and European national meteorological services to standardise AI weather forecasting workflows.
- Earth System Variable Coverage: Forecasts a broad range of variables including precipitation, temperature, ocean wind waves, snow fields, sea ice, and tropical cyclone tracks for comprehensive Earth system modelling.
- Ensemble & Probabilistic Forecasting: Supports ensemble AIFS runs that provide probabilistic guidance and uncertainty quantification, critical for high-impact weather events and operational decision-making.
Use Cases
- Meteorological researchers evaluating and benchmarking AI-based versus physics-based global weather forecast accuracy across different weather regimes and regions.
- National weather services and meteorological organisations adopting the Anemoi framework to build and operationalise their own AI weather forecasting models.
- Climate and Earth system scientists integrating ocean waves, snow, sea ice, and hydrology into machine learning forecast pipelines using open AIFS data.
- Developers and data scientists building weather-dependent applications (energy, agriculture, logistics) on top of ECMWF open data and AIFS model outputs.
- Academic institutions and students using the training-ready ERA5 dataset and open AIFS weights to study and advance machine learning methods in numerical weather prediction.
Pros
- Fully Open Access: Model weights, open data, and ERA5 training datasets are freely available, making cutting-edge AI weather forecasting accessible to the global research and developer community.
- State-of-the-Art Accuracy: AIFS demonstrates competitive and often superior performance to traditional IFS physics-based forecasts, particularly for large-scale atmospheric patterns and medium-range predictions.
- Extensible & Collaborative Framework: The Anemoi framework enables meteorological organisations to build on shared infrastructure, reducing duplication and accelerating AI adoption across national weather services.
- Comprehensive Variable Support: Goes beyond standard atmospheric variables to include ocean waves, snow, sea ice, and hydrology, advancing integrated Earth system ML modelling.
Cons
- Localised Extremes Remain Challenging: Current ML-based forecasts still struggle with localised wind extremes, storm fine structure, and fine-scale phenomena compared to high-resolution physics models.
- Requires Technical Expertise to Use: Running or fine-tuning AIFS models demands significant ML and HPC expertise; it is not designed for non-technical end users seeking simple forecast products.
- Uneven Global Performance: Improvements from AIFS may not be distributed evenly worldwide, with some regions and weather regimes benefiting less than others.
Frequently Asked Questions
What does AIFS stand for?
AIFS stands for Artificial Intelligence Forecasting System. It is ECMWF's machine learning-based global numerical weather prediction system, complementing the traditional physics-based IFS (Integrated Forecasting System).
Are AIFS model weights freely available?
Yes. ECMWF released the first AIFS model weights as open data in December 2024. They can be accessed via ECMWF's open data portal, allowing researchers and developers to run and build on the models.
What is the Anemoi framework?
Anemoi is an open-source ML framework for data-driven weather forecasting, co-developed by ECMWF and European national meteorological services. It provides shared tools for data loading, training, and inference to move AI weather models from research into operations.
What resolution does AIFS run at?
AIFS currently runs at a horizontal resolution of approximately 28 km (0.25°), providing global weather forecasts with fine spatial detail suitable for medium-range prediction.
Can I run AIFS models myself?
Yes. ECMWF has made it possible for users to run AI models using ECMWF open data. The open model weights and the Anemoi training-ready ERA5 dataset enable independent training, inference, and development of custom applications.
