Resources
Resources webinars
What’s next? Webinar
The AI Weather Quest: What’s Next? webinar took place on 16 July 2026, offering a look ahead at the competition’s extended framework and what participants can expect beyond the first forecasting year. The session notably included a step-by-step guide to submitting forecasts, visualising outputs, and self-evaluating results using the updated AI Weather Quest Python Package. Access the webinar materials:
Pre-Competition Webinar
The AI Weather Quest Testing Pre-Competition Webinars took place on 22 July 2025, and provided valuable information about the Competition Phase timeline, the evaluation system, the results display, as well as the competition outputs transparency. Access the webinar materials:
The Testing Period Launch Webinar
The AI Weather Quest Testing Period Launch Webinar took place on 7 May 2025, and provided a step-by-step demonstration on submitting forecasts, visualising outputs, and self-evaluating results using the AI Weather Quest Python Package. Access the webinar materials:
The Quest Launch Webinar
The AI Weather Quest Launch Webinars took place on 17 March 2025, and introduced the competition, its background, participation guidelines and engagement tools, followed by an extensive Q&A session. Access the webinars materials:
Model development data
Participants are encouraged to use any forecast or observational dataset to develop their AI/ML forecasting models. Below is an outline of potential data sources that may be particularly useful.
Please note that participation in the AI Weather Quest does not entitle teams to additional free or discounted access to ECMWF’s real-time operational data or licensed products. However, we warmly encourage participants to explore ECMWF’s free and open data offerings.
ERA5
Post-processed (1979 – four months before the present day, updated automatically) historical datasets of variables requested for competitive participation are freely available, including:
- Weekly means of ERA5 temperature (K) and pressure (Pa).
- Weekly accumulations of ERA5 precipitation (mm week-1).
- Daily values of observed MJO state.
- Weekly total number of tropical storm days.
The datasets are saved in annual files and can be accessed via the retrieve_training_data.py module of the AI-WQ-package Python package. Each spatial file, approximately 41 MB in size, contains weekly statistics at a daily, 1.5 degree resolution.
ECMWF Open data
The ECMWF Open data catalogue is an excellent resource for participants to initialise their already-trained AI/ML models. Specific data potentially useful for participating in the Quest includes:
- Analysis data at a 0.25° resolution through downloading initial data from high-resolution products that commence at 00 and 12 UTC.
- Forecast data at a 0.25° resolution available up to a 15-day lead time.
Comprehensive documentation outlines several methods for downloading such data, including options via multiple cloud services and a dedicated Python package.
Dynamical sub-seasonal forecast data
Participants are encouraged to apply AI/ML techniques to improve sub-seasonal forecasts derived from dynamical models. To support this, it is recommended using freely available forecast data from the following sources:
World Weather Research Programme/World Climate Research Programme sub-seasonal to seasonal (S2S) public data portal hosted by ECMWF.
Forecast data from a Subseasonal Consortium made available through an International Research Institute (IRI) database.
Predictions that are accessible through the China Meteorological Administration (CMA) S2S archiving data center.
For the ECMWF-hosted public data portal, real-time sub-seasonal forecasts and reforecast data is available from the ECMWF Data Store.
Computational resources
Whilst ECMWF cannot provide computational resources, participants are encouraged to explore external opportunities for high-performance computing (HPC) support.
Participants may choose to apply for HPC resource from computing centers. However, the HPC landscape is rapidly evolving, and demand for these resources is highly competitive.
To support participants, the AI Weather Quest organisers:
- Endeavor to assist participants from ECMWF’s Member or Co-operating States (restriction due to funding constraints) in preparing applications for European initiatives such as:
- European Weather Cloud.
- Euro HPC access calls for AI and Data-Intensive Applications.
- Calls that are released from AI factories.
- ECMWF’s special projects HPC allocation.
- Regularly share updates on available HPC opportunities through the competition’s forum.
ECMWF Learning
Participants are strongly encouraged to take advantage of ECMWF’s extensive eLearning resources to enhance their understanding and skills. Key training courses relevant to the AI Weather Quest include:
- The ECMWF Extended range forecasts: Introduction
- All modules from the Massive Open Online Course (MOOC) on Machine Learning in Weather & Climate
- Machine Learning for Earth Systems Modelling – three online courses (March to late 2026, online) – Three online courses on Machine Learning (ML) for Earth systems modelling are being developed and offered in 2026
Please note you will need an ECMWF account to access these resources. If you do not have one, please register on the ECMWF website: https://www.ecmwf.int/. This account also grants access to the competition forum, webinar registrations, and more.
Collaborative resources / networking opportunities
Tools and frameworks
- Anemoi framework (ECMWF) – To support the development and integration of machine-learning models for Earth system science, ECMWF has introduced Anemoi, a modular, open-source framework that streamlines the training, evaluation, and deployment of AI/ML models. Anemoi enables reproducible workflows, supports hybrid modelling approaches, and is being co-developed with ECMWF’s Member States and a growing network of external partners. Participants are encouraged to explore Anemoi to accelerate experimentation and model integration.
Initiatives
- WGSIP (Working Group on Subseasonal to Interdecadal Prediction) – ESMO (Earth System Modelling and Observations) activity “ML models for sub-seasonal and seasonal prediction”: Initial white paper
- WP-MIP: the Weather Prediction Model Intercomparison Project – includes sub-seasonal forecasting: White paper
Conference sessions & Events
- S2S2D Conference 2026 (7-11 September 2026, Reading) – Third International Conference on Subseasonal to Seasonal to Decadal Prediction (S2S2D)
