Predicting weather beyond two weeks remains one of the greatest challenges in meteorology. While traditional models are very good at short- and medium-range forecasting, their accuracy declines significantly at sub-seasonal timescales (15 days to 2 months ahead). This gap is crucial, as improved sub-seasonal forecasts can support decision-making in sectors such as energy, agriculture, and disaster risk management.
Recent advances in artificial intelligence (AI) and machine learning (ML) offer exciting new possibilities for improving weather forecasting, particularly at the sub-seasonal scale. AI/ML models can extract patterns from vast datasets and generate forecasts that may complement traditional physics-based approaches. To address this, ECMWF is launching the AI Weather Quest, a global competition aimed at pushing the boundaries of AI‑driven sub-seasonal forecasting.
This initiative, endorsed by the World Meteorological Organization (WMO), builds upon the foundation of the 2021 WMO Sub‑seasonal to Seasonal (S2S) AI Challenge.

A global challenge for AI and meteorology experts
The AI Weather Quest invites participants from around the world to develop and submit AI/ML‑based sub‑seasonal forecasts. The competition aims to benchmark the performance of AI‑based models, exploring how they can enhance operational forecasting.
Starting in March 2025, the competition will unfold in two phases:
- Initial Training Phase (March– August 2025): Participants will refine their models and familiarise themselves with the competition’s submission and evaluation process in a non‑competitive environment.
- Competition Phase (August 2025–at least September 2026): Participants will submit weekly, real-time forecasts over four 13‑week periods. They will be evaluated based on the Ranked Probability Skill Score (RPSS), comparing their forecasts to established benchmarks.
Forecasts will focus on three key variables: near-surface (2 m) temperature, mean sea level pressure, and precipitation. AI models must predict quintile probabilities for two lead times: days 19–25 and days 26–32, providing a probabilistic outlook crucial for decision-making in weather-sensitive industries. Forecasts will be required at a 1.5‑degree resolution, with evaluations based on ECMWF’s ERA5 reanalysis datasets.
Participants can submit up to three different AI models and provide up to 18 submissions per week (three variables across two lead times for up to three models). Models can be developed using any programming language and dataset, including observational data, reanalysis products, and existing physics-based sub-seasonal forecasts.

Open participation and recognition
The competition is open to a wide range of participants, including AI/ML researchers, meteorologists, participants from startups, large technology companies, and public forecasting institutions. No prior experience in weather forecasting is required – only a strong interest in leveraging AI to tackle real-world forecasting challenges.
ECMWF will compute and publish weekly RPSS scores, with leaderboards displaying the best-performing models. At the end of each 13‑week competition period, the top teams will be recognised in dedicated award celebrations, showcasing leading AI‑driven forecasting approaches. To ensure inclusivity and fairness, ECMWF will also highlight exceptional models from diverse organisation types and those developed with limited computational resources.
Resources for participants
To support model development, participants have access to a range of resources, including historical datasets such as ERA5. In addition, ECMWF’s Open Data Catalogue offers valuable real-time and historical forecasting products. A dedicated AI Weather Quest (AI‑WQ) Python package is available to facilitate forecast submission and evaluation, ensuring consistency and transparency throughout the competition.
ECMWF’s role and vision
As one of the world’s leading weather forecasting institutions, ECMWF is uniquely positioned to explore the integration of AI into operational meteorology. By hosting this competition, ECMWF seeks to evaluate AI/ML potential for sub‑seasonal forecasting and identify promising AI‑driven approaches. The findings from this initiative could help guide the future of operational forecasting.
Get involved
The AI Weather Quest is more than just a competition, it is a unique opportunity to contribute to the future of weather forecasting. Whether you are an AI expert, a meteorologist, or a company interested in applying machine learning to forecasting, this challenge provides a platform to test ideas, gain visibility, and make valuable connections.
Registration is now open. For more details, visit the AI Weather Quest website at https://aiweatherquest.ecmwf.int/.