Competition Structure
The AI Weather Quest kicks off in March 2025 and initially runs until August 2026. The Quest is structured into two key phases: an Initial Training Phase and a Competition Phase. Although the Quest is designed to be sequential, participants can join at the start of any period.
Initial Training Phase: 17th March 2025 – 13th August 2025
This phase is designed for participants to train and refine their forecasting models while mastering the competition tools.
During the Initial Training Phase, participants will have the opportunity to take part in a 13-week test forecasting period, the Testing JJA period (June, July, August 2025), which enables to:
- Download training data
- Submit forecast
- Self-evaluate submitted forecasts
- Engage with fellow competitors
Please note that during the Testing JJA period, verification scores and forecasts will not be displayed online.
The key engagement opportunities during the Initial Training phase include:
Quest Launch Webinar
Testing Period Launch Webinar
Testing Period Insight Webinar
These webinars provide an opportunity for participants to engage directly with organisers, ask questions, and collaborate with fellow participants. They are designed to foster discussions on challenges, progress, and results.
Join one of our two slots of the Testing Period Launch Webinar on May 7th to learn more about the Testing JJA Period and ask your questions live:
Competition Phase: 14th August 2025 – at least September 2026
The official competition begins in August 2025 and runs for at least one year. Participants will submit weekly real-time sub-seasonal forecasts using their AI models. To foster engagement, the first forecasting year is split into four 13-week competitive periods, each marked by regular opportunities for interaction and collaboration:
ECMWF will compute global Ranked Probability Skill Scores (RPSS) weekly and display top results on the AI Weather Quest website. Participants can choose to have their results anonymised for public display. At the end of each competitive period, ECMWF will calculate the average RPSS and recognise top-performing models over a 13-week period during awards webinars. At the end of the first forecasting year (September 2026), an in-person event will be organised to enable participants to present their yearly results and make valuable connections.
Participants can compete in as many 13-week competitive periods as they wish, maximising flexibility and opportunities for success. Between periods, participants can continuously improve their forecast models, enabling a real-time iterative development.
Join the Quest