Competitors in the AI Weather Quest use AI techniques to create sub-seasonal weather predictions, a forecasting time range that bridges the gap between long and short-term forecasts, that is vital for enabling regions to prepare for extreme weather events such as cyclones and cold spells.
The teams, which span 15 countries, then wait for real-life weather events unfold, to see how accurate their forecasts turned out to be.
The competition, organised by the European Centre for Medium-Range Weather Forecasts (ECMWF), is designed to increase collaboration and innovation in sub-seasonal weather predictions because it is notoriously hard to forecast in this time window, due to complex interactions in the atmospheric circulation.
Sub-seasonal forecasts, unlike seasonal forecasting, predict conditions within a specific period of a season, helping to indicate when communities will be affected. They can also refine predictions from large continental areas to much more specific regional locations than seasonal forecasts.
For example, while seasonal forecasts may indicate an increased chance of cyclones in broad regions of the Indian Ocean, sub-seasonal forecasts can predict risk at a more specific regional level, such as north-west Madagascar. Similarly, instead of predicting an elevated risk of cold conditions across all of Europe, sub-seasonal forecasts may highlight a higher likelihood of winter hazards at a country-level, e.g. France, during a particular period.
This improved level of detail helps communities get resources to the right places and to take action, such as through preparing for evacuation, reinforcing homes, or stocking up on food.
During the contest, which has attracted 42 teams in its first competitive year, participants submit sub-seasonal forecasts every week and are scored on how accurate their models turned out to be. The weekly scores are then displayed live on the AI Weather Quest website and aggregated for each 13-week competition period.
Today ECMWF reveals that the winning team of the latest period, December 2025 to February 2026 (DJF 2025/26), is MicroEnsemble.
The team, led by Microsoft, and comprised of scientists with strengths in meteorology, engineering, statistics, and AI were the most consistent performer for each weather variable across temperature, mean sea-level pressure and precipitation. Their approach uses AI technologies to post-process state-of-the-art dynamical forecasts from ECMWF.
Speaking on behalf of the team, Lester Mackey, a Senior Principal Researcher at Microsoft Research, said:
" Tackling the AI Weather Quest has been an exhilarating process and a valuable learning experience in improving probabilistic forecasts. We believe our success comes from assembling a stellar team with complementary strengths in meteorology, engineering, statistics, and AI and a common passion for building solutions that benefit society. We will continue to improve our sub-seasonal forecasting techniques and look forward to collaborating with ECMWF and other sub-seasonal AI developers ."
The leaderboard remains close at the top, with MicroEnsemble finishing just above LP, a China-based team who finished in second place for forecasting three weeks ahead and third place for forecasting four weeks ahead. ECMWF’s own team finished in third place for forecasting three weeks ahead and fourth place for forecasting four weeks ahead.
All three teams have demonstrated the rapid progress of both post-processing and purely data-driven approaches, and these will continue to evolve in the next periods of the competition as teams refine their models through this real-time benchmarking process.
Lu Peng, a Senior Engineer from Jiangsu Climate Center speaking on behalf of the LP team, said:
“ I’m very grateful to ECMWF for such a valuable opportunity to test ideas in an environment close to real-world forecasting. Our simple approach to precipitation prediction required fewer than 100 lines of additional code and runs in less than ten seconds on a normal computer without a GPU. It shows valuable experiments can happen with relatively simple tools, accessible to many. Working with participants from different backgrounds who share strong expertise and enthusiasm, allows us to work toward developing the next generation of forecasting systems .”
While most submissions come from Europe, China, and the United States, teams from Niger, Morocco, Kenya, South Africa, Peru, and South Korea have also entered, reflecting the increasingly global reach of AI/ML approaches to forecasting.
Kenyan-based team Fahamu have consistently submitted forecasts and led the way in using Anemoi technologies to enable operational sub-seasonal forecasting in a developing country.
Speaking on behalf of the team, Nishadh Kalladath, a Data Scientist and Machine Learning expert at the IGAD Climate Prediction and Applications Centre (ICPAC) , said:
“ AI Weather Quest provides a unique opportunity for operational forecasting centres, researchers, and practitioners working on weather and climate to explore jointly, how emerging AI methods can complement and extend traditional weather prediction systems. This collaboration is essential if we want AI-driven weather and climate forecasting to become part of operational early warning systems that benefit communities on the ground. For our team in East Africa, reliable sub-seasonal forecasts are essential for improving early warning systems and supporting anticipatory action for hazards such as droughts and floods. AI Weather Quest allows us to test how AI-based ensemble prediction systems can be translated into actionable information for decision-makers. ”
ECMWF’s own team in the contest applied the Artificial Intelligence/Integrated Forecasting System (AIFS) and is currently the best ranked purely data driven model, i.e. not using outputs from traditional physics-based weather models as inputs.
Jakob Schloer, Scientist for data-driven sub-seasonal forecasting at ECMWF and lead of the AIFS-team for the Weather Quest, said:
“ The AI Weather Quest gives us a great opportunity to put different versions of our AIFS model to the test in real time already at an early development stage. This is both exciting and genuinely fun for the team. We're happy that AIFS has established itself as the best-performing purely data-driven model in the competition. But what makes it especially valuable is the chance to see how other teams are tackling the same challenge — particularly those combining traditional numerical methods with machine learning. Ultimately, we all learn from each other, and that pushes the whole field forward. ”
The competition, funded by the European Union through the Destination Earth initiative, and endorsed as a WMO Integrated Processing and Prediction System (WIPPS) pilot project of the World Meteorological Organization (WMO) is now at the midway point of its first edition. Olga Loegel, User Outreach and Engagement Associate at ECMWF, and lead of the AI Weather Quest organisation, concluded:
“ The AI Weather Quest is not only a transparent benchmark for evaluating how artificial intelligence performs for sub-seasonal weather prediction; it is also a global learning framework. Our collaborative approach of bringing together researchers, meteorological services, and industry from ECMWF Member and Co-operating States and the international community, creates a shared space to exchange insights on where and how AI can improve forecasts. It is key to building robust scientific evidence, strengthening trust in new methods, and ultimately improving forecasts that support decision-making worldwide. Everyone in the competition, regardless of their position, is contributing to this aim. ”
Find out more in our latest blog post .
On Thursday 19 March a webinar will feature some of the top contenders, who will present their results. You can sign up here: https://events.ecmwf.int/event/486/
For more information, how to get involved, or to stay up to date with the leaderboards, see: https://aiweatherquest.ecmwf.int/ .